diff --git a/CHANGELOG.md b/CHANGELOG.md
index b6165b4..054bf18 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,7 @@ Change Log
 Unreleased
 ----------
 - Support for model constraints
+- Support for data skipping indexes
 
 v2.0.1
 ------
diff --git a/docs/class_reference.md b/docs/class_reference.md
index 616863f..bb4b099 100644
--- a/docs/class_reference.md
+++ b/docs/class_reference.md
@@ -731,16 +731,84 @@ Defines a model constraint.
 #### Constraint(expr)
 
 
-Initializer. Requires an expression that ClickHouse will verify when inserting data.
+Initializer. Expects an expression that ClickHouse will verify when inserting data.
 
 
 #### create_table_sql()
 
 
-Returns the SQL statement for defining this constraint on table creation.
+Returns the SQL statement for defining this constraint during table creation.
 
 
-#### str()
+### Index
+
+
+Defines a data-skipping index.
+
+#### Index(expr, type, granularity)
+
+
+Initializer.
+
+- `expr` - a column, expression, or tuple of columns and expressions to index.
+- `type` - the index type. Use one of the following methods to specify the type:
+  `Index.minmax`, `Index.set`, `Index.ngrambf_v1`, `Index.tokenbf_v1` or `Index.bloom_filter`.
+- `granularity` - index block size (number of multiples of the `index_granularity` defined by the engine).
+
+
+#### bloom_filter()
+
+
+An index that stores a Bloom filter containing values of the index expression.
+
+- `false_positive` - the probability (between 0 and 1) of receiving a false positive
+  response from the filter
+
+
+#### create_table_sql()
+
+
+Returns the SQL statement for defining this index during table creation.
+
+
+#### minmax()
+
+
+An index that stores extremes of the specified expression (if the expression is tuple, then it stores
+extremes for each element of tuple). The stored info is used for skipping blocks of data like the primary key.
+
+
+#### ngrambf_v1(size_of_bloom_filter_in_bytes, number_of_hash_functions, random_seed)
+
+
+An index that stores a Bloom filter containing all ngrams from a block of data.
+Works only with strings. Can be used for optimization of equals, like and in expressions.
+
+- `n` — ngram size
+- `size_of_bloom_filter_in_bytes` — Bloom filter size in bytes (you can use large values here,
+   for example 256 or 512, because it can be compressed well).
+- `number_of_hash_functions` — The number of hash functions used in the Bloom filter.
+- `random_seed` — The seed for Bloom filter hash functions.
+
+
+#### set()
+
+
+An index that stores unique values of the specified expression (no more than max_rows rows,
+or unlimited if max_rows=0). Uses the values to check if the WHERE expression is not satisfiable
+on a block of data.
+
+
+#### tokenbf_v1(number_of_hash_functions, random_seed)
+
+
+An index that stores a Bloom filter containing string tokens. Tokens are sequences
+separated by non-alphanumeric characters.
+
+- `size_of_bloom_filter_in_bytes` — Bloom filter size in bytes (you can use large values here,
+   for example 256 or 512, because it can be compressed well).
+- `number_of_hash_functions` — The number of hash functions used in the Bloom filter.
+- `random_seed` — The seed for Bloom filter hash functions.
 
 
 infi.clickhouse_orm.fields
diff --git a/docs/models_and_databases.md b/docs/models_and_databases.md
index 928489d..28ab6ad 100644
--- a/docs/models_and_databases.md
+++ b/docs/models_and_databases.md
@@ -9,17 +9,18 @@ Defining Models
 ---------------
 
 Models are defined in a way reminiscent of Django's ORM, by subclassing `Model`:
+```python
+from infi.clickhouse_orm import Model, StringField, DateField, Float32Field, MergeTree
 
-    from infi.clickhouse_orm import Model, StringField, DateField, Float32Field, MergeTree
+class Person(Model):
 
-    class Person(Model):
+    first_name = StringField()
+    last_name = StringField()
+    birthday = DateField()
+    height = Float32Field()
 
-        first_name = StringField()
-        last_name = StringField()
-        birthday = DateField()
-        height = Float32Field()
-
-        engine = MergeTree('birthday', ('first_name', 'last_name', 'birthday'))
+    engine = MergeTree('birthday', ('first_name', 'last_name', 'birthday'))
+```
 
 The columns in the database table are represented by model fields. Each field has a type, which matches the type of the corresponding database column. All the supported fields types are listed [here](field_types.md).
 
@@ -66,32 +67,45 @@ For additional details see [here](field_options.md).
 ### Table Names
 
 The table name used for the model is its class name, converted to lowercase. To override the default name, implement the `table_name` method:
+```python
+class Person(Model):
 
-    class Person(Model):
+    ...
 
-        ...
-
-        @classmethod
-        def table_name(cls):
-            return 'people'
+    @classmethod
+    def table_name(cls):
+        return 'people'
+```
 
 ### Model Constraints
 
 It is possible to define constraints which ClickHouse verifies when data is inserted. Trying to insert invalid records will raise a `ServerError`. Each constraint has a name and an expression to validate. For example:
+```python
+class Person(Model):
 
-    from infi.clickhouse_orm import Model, Constraint, F, StringField, DateField, Float32Field, MergeTree
+    ...
 
-    class Person(Model):
+    # Ensure that the birthday is not a future date
+    birthday_is_in_the_past = Constraint(birthday <= F.today())
+```
 
-        first_name = StringField()
-        last_name = StringField()
-        birthday = DateField()
-        height = Float32Field()
+### Data Skipping Indexes
 
-        # Ensure that the birthday is not a future date
-        birthday_is_in_the_past = Constraint(birthday <= F.today())
+Models that use an engine from the `MergeTree` family can define additional indexes over one or more columns or expressions. These indexes are used in SELECT queries for reducing the amount of data to read from the disk by skipping big blocks of data that do not satisfy the query's conditions.
 
-        engine = MergeTree('birthday', ('first_name', 'last_name', 'birthday'))
+For example:
+```python
+class Person(Model):
+
+    ...
+
+    # A minmax index that can help find people taller or shorter than some height
+    height_index = Index(height, type=Index.minmax(), granularity=2)
+
+    # A trigram index that can help find substrings inside people names
+    names_index = Index((F.lower(first_name), F.lower(last_name)),
+                        type=Index.ngrambf_v1(3, 256, 2, 0), granularity=1)
+```
 
 
 Using Models
diff --git a/docs/toc.md b/docs/toc.md
index 6dd5da0..36a20e7 100644
--- a/docs/toc.md
+++ b/docs/toc.md
@@ -11,6 +11,7 @@
          * [Alias fields](models_and_databases.md#alias-fields)
          * [Table Names](models_and_databases.md#table-names)
          * [Model Constraints](models_and_databases.md#model-constraints)
+         * [Data Skipping Indexes](models_and_databases.md#data-skipping-indexes)
       * [Using Models](models_and_databases.md#using-models)
       * [Inserting to the Database](models_and_databases.md#inserting-to-the-database)
       * [Reading from the Database](models_and_databases.md#reading-from-the-database)
@@ -85,6 +86,7 @@
          * [MergeModel](class_reference.md#mergemodel)
          * [DistributedModel](class_reference.md#distributedmodel)
          * [Constraint](class_reference.md#constraint)
+         * [Index](class_reference.md#index)
       * [infi.clickhouse_orm.fields](class_reference.md#inficlickhouse_ormfields)
          * [ArrayField](class_reference.md#arrayfield)
          * [BaseEnumField](class_reference.md#baseenumfield)
diff --git a/scripts/generate_ref.py b/scripts/generate_ref.py
index 8dc0477..22850ab 100644
--- a/scripts/generate_ref.py
+++ b/scripts/generate_ref.py
@@ -132,7 +132,7 @@ if __name__ == '__main__':
     print('===============')
     print()
     module_doc([database.Database, database.DatabaseException])
-    module_doc([models.Model, models.BufferModel, models.MergeModel, models.DistributedModel, models.Constraint])
+    module_doc([models.Model, models.BufferModel, models.MergeModel, models.DistributedModel, models.Constraint, models.Index])
     module_doc(sorted([fields.Field] + all_subclasses(fields.Field), key=lambda x: x.__name__), False)
     module_doc([engines.Engine] + all_subclasses(engines.Engine), False)
     module_doc([query.QuerySet, query.AggregateQuerySet, query.Q])
diff --git a/src/infi/clickhouse_orm/funcs.py b/src/infi/clickhouse_orm/funcs.py
index 8d59528..080c3d7 100644
--- a/src/infi/clickhouse_orm/funcs.py
+++ b/src/infi/clickhouse_orm/funcs.py
@@ -1,9 +1,8 @@
-from datetime import date, datetime, tzinfo, timedelta
 from functools import wraps
 from inspect import signature, Parameter
 from types import FunctionType
 
-from .utils import is_iterable, comma_join, NO_VALUE
+from .utils import is_iterable, comma_join, NO_VALUE, arg_to_sql
 from .query import Cond, QuerySet
 
 
@@ -263,43 +262,9 @@ class F(Cond, FunctionOperatorsMixin, metaclass=FMeta):
         else:
             prefix = self.name
             sep = ', '
-        arg_strs = (F._arg_to_sql(arg) for arg in self.args if arg != NO_VALUE)
+        arg_strs = (arg_to_sql(arg) for arg in self.args if arg != NO_VALUE)
         return prefix + '(' + sep.join(arg_strs) + ')'
 
-    @staticmethod
-    def _arg_to_sql(arg):
-        """
-        Converts a function argument to SQL string according to its type.
-        Supports functions, model fields, strings, dates, datetimes, timedeltas, booleans,
-        None, numbers, timezones, arrays/iterables.
-        """
-        from .fields import Field, StringField, DateTimeField, DateField
-        if isinstance(arg, F):
-            return arg.to_sql()
-        if isinstance(arg, Field):
-            return "`%s`" % arg
-        if isinstance(arg, str):
-            return StringField().to_db_string(arg)
-        if isinstance(arg, datetime):
-            return "toDateTime(%s)" % DateTimeField().to_db_string(arg)
-        if isinstance(arg, date):
-            return "toDate('%s')" % arg.isoformat()
-        if isinstance(arg, timedelta):
-            return "toIntervalSecond(%d)" % int(arg.total_seconds())
-        if isinstance(arg, bool):
-            return str(int(arg))
-        if isinstance(arg, tzinfo):
-            return StringField().to_db_string(arg.tzname(None))
-        if arg is None:
-            return 'NULL'
-        if isinstance(arg, QuerySet):
-            return "(%s)" % arg
-        if isinstance(arg, tuple):
-            return '(' + comma_join(F._arg_to_sql(x) for x in arg) + ')'
-        if is_iterable(arg):
-            return '[' + comma_join(F._arg_to_sql(x) for x in arg) + ']'
-        return str(arg)
-
     # Arithmetic functions
 
     @staticmethod
diff --git a/src/infi/clickhouse_orm/migrations.py b/src/infi/clickhouse_orm/migrations.py
index 5361b2b..0ff3f59 100644
--- a/src/infi/clickhouse_orm/migrations.py
+++ b/src/infi/clickhouse_orm/migrations.py
@@ -26,12 +26,13 @@ class ModelOperation(Operation):
         Initializer.
         '''
         self.model_class = model_class
+        self.table_name = model_class.table_name()
 
     def _alter_table(self, database, cmd):
         '''
         Utility for running ALTER TABLE commands.
         '''
-        cmd = "ALTER TABLE $db.`%s` %s" % (self.model_class.table_name(), cmd)
+        cmd = "ALTER TABLE $db.`%s` %s" % (self.table_name, cmd)
         logger.debug(cmd)
         database.raw(cmd)
 
@@ -42,7 +43,7 @@ class CreateTable(ModelOperation):
     '''
 
     def apply(self, database):
-        logger.info('    Create table %s', self.model_class.table_name())
+        logger.info('    Create table %s', self.table_name)
         if issubclass(self.model_class, BufferModel):
             database.create_table(self.model_class.engine.main_model)
         database.create_table(self.model_class)
@@ -59,11 +60,11 @@ class AlterTable(ModelOperation):
     '''
 
     def _get_table_fields(self, database):
-        query = "DESC `%s`.`%s`" % (database.db_name, self.model_class.table_name())
+        query = "DESC `%s`.`%s`" % (database.db_name, self.table_name)
         return [(row.name, row.type) for row in database.select(query)]
 
     def apply(self, database):
-        logger.info('    Alter table %s', self.model_class.table_name())
+        logger.info('    Alter table %s', self.table_name)
 
         # Note that MATERIALIZED and ALIAS fields are always at the end of the DESC,
         # ADD COLUMN ... AFTER doesn't affect it
@@ -131,7 +132,7 @@ class DropTable(ModelOperation):
     '''
 
     def apply(self, database):
-        logger.info('    Drop table %s', self.model_class.table_name())
+        logger.info('    Drop table %s', self.table_name)
         database.drop_table(self.model_class)
 
 
@@ -144,7 +145,7 @@ class AlterConstraints(ModelOperation):
     '''
 
     def apply(self, database):
-        logger.info('    Alter constraints for %s', self.model_class.table_name())
+        logger.info('    Alter constraints for %s', self.table_name)
         existing = self._get_constraint_names(database)
         # Go over constraints in the model
         for constraint in self.model_class._constraints.values():
@@ -164,8 +165,56 @@ class AlterConstraints(ModelOperation):
         Returns a set containing the names of existing constraints in the table.
         '''
         import re
-        create_table_sql = database.raw('SHOW CREATE TABLE $db.`%s`' % self.model_class.table_name())
-        matches = re.findall(r'\sCONSTRAINT\s+`?(.+?)`?\s+CHECK\s', create_table_sql, flags=re.IGNORECASE)
+        table_def = database.raw('SHOW CREATE TABLE $db.`%s`' % self.table_name)
+        matches = re.findall(r'\sCONSTRAINT\s+`?(.+?)`?\s+CHECK\s', table_def)
+        return set(matches)
+
+
+class AlterIndexes(ModelOperation):
+    '''
+    A migration operation that adds new indexes from the model to the database
+    table, and drops obsolete ones. Indexes are identified by their names, so
+    a change in an existing index will not be detected unless its name was changed too.
+    '''
+
+    def __init__(self, model_class, reindex=False):
+        '''
+        Initializer.
+        By default ClickHouse does not build indexes over existing data, only for
+        new data. Passing `reindex=True` will run `OPTIMIZE TABLE` in order to build
+        the indexes over the existing data.
+        '''
+        super().__init__(model_class)
+        self.reindex = reindex
+
+    def apply(self, database):
+        logger.info('    Alter indexes for %s', self.table_name)
+        existing = self._get_index_names(database)
+        logger.info(existing)
+        # Go over indexes in the model
+        for index in self.model_class._indexes.values():
+            # Check if it's a new index
+            if index.name not in existing:
+                logger.info('        Add index %s', index.name)
+                self._alter_table(database, 'ADD %s' % index.create_table_sql())
+            else:
+                existing.remove(index.name)
+        # Remaining indexes in `existing` are obsolete
+        for name in existing:
+            logger.info('        Drop index %s', name)
+            self._alter_table(database, 'DROP INDEX `%s`' % name)
+        # Reindex
+        if self.reindex:
+            logger.info('        Build indexes on table')
+            self.database.raw('OPTIMIZE TABLE $db.`%s` FINAL' % self.table_name)
+
+    def _get_index_names(self, database):
+        '''
+        Returns a set containing the names of existing indexes in the table.
+        '''
+        import re
+        table_def = database.raw('SHOW CREATE TABLE $db.`%s`' % self.table_name)
+        matches = re.findall(r'\sINDEX\s+`?(.+?)`?\s+', table_def)
         return set(matches)
 
 
diff --git a/src/infi/clickhouse_orm/models.py b/src/infi/clickhouse_orm/models.py
index 506185b..cc535a1 100644
--- a/src/infi/clickhouse_orm/models.py
+++ b/src/infi/clickhouse_orm/models.py
@@ -7,7 +7,7 @@ from logging import getLogger
 import pytz
 
 from .fields import Field, StringField
-from .utils import parse_tsv, NO_VALUE, get_subclass_names
+from .utils import parse_tsv, NO_VALUE, get_subclass_names, arg_to_sql
 from .query import QuerySet
 from .funcs import F
 from .engines import Merge, Distributed
@@ -16,7 +16,7 @@ logger = getLogger('clickhouse_orm')
 
 
 
-class Constraint():
+class Constraint:
     '''
     Defines a model constraint.
     '''
@@ -34,10 +34,89 @@ class Constraint():
         '''
         Returns the SQL statement for defining this constraint during table creation.
         '''
-        return 'CONSTRAINT `%s` CHECK %s' % (self.name, self.expr)
+        return 'CONSTRAINT `%s` CHECK %s' % (self.name, arg_to_sql(self.expr))
 
-    def str(self):
-        return self.create_table_sql()
+
+class Index:
+    '''
+    Defines a data-skipping index.
+    '''
+
+    name   = None # this is set by the parent model
+    parent = None # this is set by the parent model
+
+    def __init__(self, expr, type, granularity):
+        '''
+        Initializer.
+
+        - `expr` - a column, expression, or tuple of columns and expressions to index.
+        - `type` - the index type. Use one of the following methods to specify the type:
+          `Index.minmax`, `Index.set`, `Index.ngrambf_v1`, `Index.tokenbf_v1` or `Index.bloom_filter`.
+        - `granularity` - index block size (number of multiples of the `index_granularity` defined by the engine).
+        '''
+        self.expr = expr
+        self.type = type
+        self.granularity = granularity
+
+    def create_table_sql(self):
+        '''
+        Returns the SQL statement for defining this index during table creation.
+        '''
+        return 'INDEX `%s` %s TYPE %s GRANULARITY %d' % (self.name, arg_to_sql(self.expr), self.type, self.granularity)
+
+    @staticmethod
+    def minmax():
+        '''
+        An index that stores extremes of the specified expression (if the expression is tuple, then it stores
+        extremes for each element of tuple). The stored info is used for skipping blocks of data like the primary key.
+        '''
+        return 'minmax'
+
+    @staticmethod
+    def set(max_rows):
+        '''
+        An index that stores unique values of the specified expression (no more than max_rows rows,
+        or unlimited if max_rows=0). Uses the values to check if the WHERE expression is not satisfiable
+        on a block of data.
+        '''
+        return 'set(%d)' % max_rows
+
+    @staticmethod
+    def ngrambf_v1(n, size_of_bloom_filter_in_bytes, number_of_hash_functions, random_seed):
+        '''
+        An index that stores a Bloom filter containing all ngrams from a block of data.
+        Works only with strings. Can be used for optimization of equals, like and in expressions.
+
+        - `n` — ngram size
+        - `size_of_bloom_filter_in_bytes` — Bloom filter size in bytes (you can use large values here,
+           for example 256 or 512, because it can be compressed well).
+        - `number_of_hash_functions` — The number of hash functions used in the Bloom filter.
+        - `random_seed` — The seed for Bloom filter hash functions.
+        '''
+        return 'ngrambf_v1(%d, %d, %d, %d)' % (n, size_of_bloom_filter_in_bytes, number_of_hash_functions, random_seed)
+
+    @staticmethod
+    def tokenbf_v1(size_of_bloom_filter_in_bytes, number_of_hash_functions, random_seed):
+        '''
+        An index that stores a Bloom filter containing string tokens. Tokens are sequences
+        separated by non-alphanumeric characters.
+
+        - `size_of_bloom_filter_in_bytes` — Bloom filter size in bytes (you can use large values here,
+           for example 256 or 512, because it can be compressed well).
+        - `number_of_hash_functions` — The number of hash functions used in the Bloom filter.
+        - `random_seed` — The seed for Bloom filter hash functions.
+        '''
+        return 'tokenbf_v1(%d, %d, %d)' % (size_of_bloom_filter_in_bytes, number_of_hash_functions, random_seed)
+
+    @staticmethod
+    def bloom_filter(false_positive=0.025):
+        '''
+        An index that stores a Bloom filter containing values of the index expression.
+
+        - `false_positive` - the probability (between 0 and 1) of receiving a false positive
+          response from the filter
+        '''
+        return 'bloom_filter(%f)' % false_positive
 
 
 class ModelBase(type):
@@ -48,21 +127,29 @@ class ModelBase(type):
     ad_hoc_model_cache = {}
 
     def __new__(cls, name, bases, attrs):
-        # Collect fields and constraints from parent classes
-        fields = dict()
-        constraints = dict()
+
+        # Collect fields, constraints and indexes from parent classes
+        fields = {}
+        constraints = {}
+        indexes = {}
         for base in bases:
             if isinstance(base, ModelBase):
                 fields.update(base._fields)
                 constraints.update(base._constraints)
+                indexes.update(base._indexes)
 
-        # Build a list of (name, field) tuples, in the order they were listed in the class
-        fields.update({n: f for n, f in attrs.items() if isinstance(f, Field)})
+        # Add fields, constraints and indexes from this class
+        for n, obj in attrs.items():
+            if isinstance(obj, Field):
+                fields[n] = obj
+            elif isinstance(obj, Constraint):
+                constraints[n] = obj
+            elif isinstance(obj, Index):
+                indexes[n] = obj
+
+        # Convert fields to a list of (name, field) tuples, in the order they were listed in the class
         fields = sorted(fields.items(), key=lambda item: item[1].creation_counter)
 
-        # Build a list of constraints
-        constraints.update({n: c for n, c in attrs.items() if isinstance(c, Constraint)})
-
         # Build a dictionary of default values
         defaults = {}
         has_funcs_as_defaults = False
@@ -75,18 +162,20 @@ class ModelBase(type):
             else:
                 defaults[n] = f.to_python(f.default, pytz.UTC)
 
+        # Create the model class
         attrs = dict(
             attrs,
             _fields=OrderedDict(fields),
             _constraints=constraints,
+            _indexes=indexes,
             _writable_fields=OrderedDict([f for f in fields if not f[1].readonly]),
             _defaults=defaults,
             _has_funcs_as_defaults=has_funcs_as_defaults
         )
         model = super(ModelBase, cls).__new__(cls, str(name), bases, attrs)
 
-        # Let each field and constraint know its parent and its own name
-        for n, obj in chain(fields, constraints.items()):
+        # Let each field, constraint and index know its parent and its own name
+        for n, obj in chain(fields, constraints.items(), indexes.items()):
             setattr(obj, 'parent', model)
             setattr(obj, 'name', n)
 
@@ -255,24 +344,22 @@ class Model(metaclass=ModelBase):
         Returns the SQL statement for creating a table for this model.
         '''
         parts = ['CREATE TABLE IF NOT EXISTS `%s`.`%s` (' % (db.db_name, cls.table_name())]
-        cols = []
+        # Fields
+        items = []
         for name, field in cls.fields().items():
-            cols.append('    %s %s' % (name, field.get_sql(db=db)))
-        parts.append(',\n'.join(cols))
-        parts.append(cls._constraints_sql())
+            items.append('    %s %s' % (name, field.get_sql(db=db)))
+        # Constraints
+        for c in cls._constraints.values():
+            items.append('    %s' % c.create_table_sql())
+        # Indexes
+        for i in cls._indexes.values():
+            items.append('    %s' % i.create_table_sql())
+        parts.append(',\n'.join(items))
+        # Engine
         parts.append(')')
         parts.append('ENGINE = ' + cls.engine.create_table_sql(db))
         return '\n'.join(parts)
 
-    @classmethod
-    def _constraints_sql(cls):
-        '''
-        Returns this model's contraints as SQL.
-        '''
-        if not cls._constraints:
-            return ''
-        return ',' + ',\n'.join(c.create_table_sql() for c in cls._constraints.values())
-
     @classmethod
     def drop_table_sql(cls, db):
         '''
@@ -420,7 +507,6 @@ class MergeModel(Model):
             if name != '_table':
                 cols.append('    %s %s' % (name, field.get_sql(db=db)))
         parts.append(',\n'.join(cols))
-        parts.append(cls._constraints_sql())
         parts.append(')')
         parts.append('ENGINE = ' + cls.engine.create_table_sql(db))
         return '\n'.join(parts)
@@ -510,4 +596,4 @@ class DistributedModel(Model):
 
 
 # Expose only relevant classes in import *
-__all__ = get_subclass_names(locals(), (Model, Constraint))
+__all__ = get_subclass_names(locals(), (Model, Constraint, Index))
diff --git a/src/infi/clickhouse_orm/utils.py b/src/infi/clickhouse_orm/utils.py
index a487bfe..c0d0325 100644
--- a/src/infi/clickhouse_orm/utils.py
+++ b/src/infi/clickhouse_orm/utils.py
@@ -1,6 +1,6 @@
-from __future__ import unicode_literals
 import codecs
 import re
+from datetime import date, datetime, tzinfo, timedelta
 
 
 SPECIAL_CHARS = {
@@ -42,6 +42,40 @@ def string_or_func(obj):
     return obj.to_sql() if hasattr(obj, 'to_sql') else obj
 
 
+def arg_to_sql(arg):
+    """
+    Converts a function argument to SQL string according to its type.
+    Supports functions, model fields, strings, dates, datetimes, timedeltas, booleans,
+    None, numbers, timezones, arrays/iterables.
+    """
+    from infi.clickhouse_orm import Field, StringField, DateTimeField, DateField, F, QuerySet
+    if isinstance(arg, F):
+        return arg.to_sql()
+    if isinstance(arg, Field):
+        return "`%s`" % arg
+    if isinstance(arg, str):
+        return StringField().to_db_string(arg)
+    if isinstance(arg, datetime):
+        return "toDateTime(%s)" % DateTimeField().to_db_string(arg)
+    if isinstance(arg, date):
+        return "toDate('%s')" % arg.isoformat()
+    if isinstance(arg, timedelta):
+        return "toIntervalSecond(%d)" % int(arg.total_seconds())
+    if isinstance(arg, bool):
+        return str(int(arg))
+    if isinstance(arg, tzinfo):
+        return StringField().to_db_string(arg.tzname(None))
+    if arg is None:
+        return 'NULL'
+    if isinstance(arg, QuerySet):
+        return "(%s)" % arg
+    if isinstance(arg, tuple):
+        return '(' + comma_join(arg_to_sql(x) for x in arg) + ')'
+    if is_iterable(arg):
+        return '[' + comma_join(arg_to_sql(x) for x in arg) + ']'
+    return str(arg)
+
+
 def parse_tsv(line):
     if isinstance(line, bytes):
         line = line.decode()
diff --git a/tests/sample_migrations/0018.py b/tests/sample_migrations/0018.py
new file mode 100644
index 0000000..c34c137
--- /dev/null
+++ b/tests/sample_migrations/0018.py
@@ -0,0 +1,6 @@
+from infi.clickhouse_orm import migrations
+from ..test_migrations import *
+
+operations = [
+    migrations.CreateTable(ModelWithIndex)
+]
diff --git a/tests/sample_migrations/0019.py b/tests/sample_migrations/0019.py
new file mode 100644
index 0000000..67ba244
--- /dev/null
+++ b/tests/sample_migrations/0019.py
@@ -0,0 +1,6 @@
+from infi.clickhouse_orm import migrations
+from ..test_migrations import *
+
+operations = [
+    migrations.AlterIndexes(ModelWithIndex2, reindex=True)
+]
diff --git a/tests/test_indexes.py b/tests/test_indexes.py
new file mode 100644
index 0000000..2507e92
--- /dev/null
+++ b/tests/test_indexes.py
@@ -0,0 +1,32 @@
+import unittest
+
+from infi.clickhouse_orm import *
+
+
+class IndexesTest(unittest.TestCase):
+
+    def setUp(self):
+        self.database = Database('test-db', log_statements=True)
+        if self.database.server_version < (19, 3, 3):
+            raise unittest.SkipTest('ClickHouse version too old')
+
+    def tearDown(self):
+        self.database.drop_database()
+
+    def test_all_index_types(self):
+        self.database.create_table(ModelWithIndexes)
+
+
+class ModelWithIndexes(Model):
+
+    date = DateField()
+    f1 = Int32Field()
+    f2 = StringField()
+
+    i1 = Index(f1, type=Index.minmax(), granularity=1)
+    i2 = Index(f1, type=Index.set(1000), granularity=2)
+    i3 = Index(f2, type=Index.ngrambf_v1(3, 256, 2, 0), granularity=1)
+    i4 = Index(F.lower(f2), type=Index.tokenbf_v1(256, 2, 0), granularity=2)
+    i5 = Index((F.toQuarter(date), f2), type=Index.bloom_filter(), granularity=3)
+
+    engine = MergeTree('date', ('date',))
diff --git a/tests/test_migrations.py b/tests/test_migrations.py
index e8d7776..61b06b8 100644
--- a/tests/test_migrations.py
+++ b/tests/test_migrations.py
@@ -2,7 +2,7 @@ from __future__ import unicode_literals
 import unittest
 
 from infi.clickhouse_orm.database import Database, ServerError
-from infi.clickhouse_orm.models import Model, BufferModel, Constraint
+from infi.clickhouse_orm.models import Model, BufferModel, Constraint, Index
 from infi.clickhouse_orm.fields import *
 from infi.clickhouse_orm.engines import *
 from infi.clickhouse_orm.migrations import MigrationHistory
@@ -27,55 +27,58 @@ class MigrationsTestCase(unittest.TestCase):
     def tearDown(self):
         self.database.drop_database()
 
-    def tableExists(self, model_class):
+    def table_exists(self, model_class):
         query = "EXISTS TABLE $db.`%s`" % model_class.table_name()
         return next(self.database.select(query)).result == 1
 
-    def getTableFields(self, model_class):
+    def get_table_fields(self, model_class):
         query = "DESC `%s`.`%s`" % (self.database.db_name, model_class.table_name())
         return [(row.name, row.type) for row in self.database.select(query)]
 
+    def get_table_def(self, model_class):
+        return self.database.raw('SHOW CREATE TABLE $db.`%s`' % self.table_name)
+
     def test_migrations(self):
         # Creation and deletion of table
         self.database.migrate('tests.sample_migrations', 1)
-        self.assertTrue(self.tableExists(Model1))
+        self.assertTrue(self.table_exists(Model1))
         self.database.migrate('tests.sample_migrations', 2)
-        self.assertFalse(self.tableExists(Model1))
+        self.assertFalse(self.table_exists(Model1))
         self.database.migrate('tests.sample_migrations', 3)
-        self.assertTrue(self.tableExists(Model1))
+        self.assertTrue(self.table_exists(Model1))
         # Adding, removing and altering simple fields
-        self.assertEqual(self.getTableFields(Model1), [('date', 'Date'), ('f1', 'Int32'), ('f2', 'String')])
+        self.assertEqual(self.get_table_fields(Model1), [('date', 'Date'), ('f1', 'Int32'), ('f2', 'String')])
         self.database.migrate('tests.sample_migrations', 4)
-        self.assertEqual(self.getTableFields(Model2), [('date', 'Date'), ('f1', 'Int32'), ('f3', 'Float32'), ('f2', 'String'), ('f4', 'String'), ('f5', 'Array(UInt64)')])
+        self.assertEqual(self.get_table_fields(Model2), [('date', 'Date'), ('f1', 'Int32'), ('f3', 'Float32'), ('f2', 'String'), ('f4', 'String'), ('f5', 'Array(UInt64)')])
         self.database.migrate('tests.sample_migrations', 5)
-        self.assertEqual(self.getTableFields(Model3), [('date', 'Date'), ('f1', 'Int64'), ('f3', 'Float64'), ('f4', 'String')])
+        self.assertEqual(self.get_table_fields(Model3), [('date', 'Date'), ('f1', 'Int64'), ('f3', 'Float64'), ('f4', 'String')])
         # Altering enum fields
         self.database.migrate('tests.sample_migrations', 6)
-        self.assertTrue(self.tableExists(EnumModel1))
-        self.assertEqual(self.getTableFields(EnumModel1),
+        self.assertTrue(self.table_exists(EnumModel1))
+        self.assertEqual(self.get_table_fields(EnumModel1),
                           [('date', 'Date'), ('f1', "Enum8('dog' = 1, 'cat' = 2, 'cow' = 3)")])
         self.database.migrate('tests.sample_migrations', 7)
-        self.assertTrue(self.tableExists(EnumModel1))
-        self.assertEqual(self.getTableFields(EnumModel2),
+        self.assertTrue(self.table_exists(EnumModel1))
+        self.assertEqual(self.get_table_fields(EnumModel2),
                           [('date', 'Date'), ('f1', "Enum16('dog' = 1, 'cat' = 2, 'horse' = 3, 'pig' = 4)")])
         # Materialized fields and alias fields
         self.database.migrate('tests.sample_migrations', 8)
-        self.assertTrue(self.tableExists(MaterializedModel))
-        self.assertEqual(self.getTableFields(MaterializedModel),
+        self.assertTrue(self.table_exists(MaterializedModel))
+        self.assertEqual(self.get_table_fields(MaterializedModel),
                           [('date_time', "DateTime"), ('date', 'Date')])
         self.database.migrate('tests.sample_migrations', 9)
-        self.assertTrue(self.tableExists(AliasModel))
-        self.assertEqual(self.getTableFields(AliasModel),
+        self.assertTrue(self.table_exists(AliasModel))
+        self.assertEqual(self.get_table_fields(AliasModel),
                           [('date', 'Date'), ('date_alias', "Date")])
         # Buffer models creation and alteration
         self.database.migrate('tests.sample_migrations', 10)
-        self.assertTrue(self.tableExists(Model4))
-        self.assertTrue(self.tableExists(Model4Buffer))
-        self.assertEqual(self.getTableFields(Model4), [('date', 'Date'), ('f1', 'Int32'), ('f2', 'String')])
-        self.assertEqual(self.getTableFields(Model4Buffer), [('date', 'Date'), ('f1', 'Int32'), ('f2', 'String')])
+        self.assertTrue(self.table_exists(Model4))
+        self.assertTrue(self.table_exists(Model4Buffer))
+        self.assertEqual(self.get_table_fields(Model4), [('date', 'Date'), ('f1', 'Int32'), ('f2', 'String')])
+        self.assertEqual(self.get_table_fields(Model4Buffer), [('date', 'Date'), ('f1', 'Int32'), ('f2', 'String')])
         self.database.migrate('tests.sample_migrations', 11)
-        self.assertEqual(self.getTableFields(Model4), [('date', 'Date'), ('f3', 'DateTime'), ('f2', 'String')])
-        self.assertEqual(self.getTableFields(Model4Buffer), [('date', 'Date'), ('f3', 'DateTime'), ('f2', 'String')])
+        self.assertEqual(self.get_table_fields(Model4), [('date', 'Date'), ('f3', 'DateTime'), ('f2', 'String')])
+        self.assertEqual(self.get_table_fields(Model4Buffer), [('date', 'Date'), ('f3', 'DateTime'), ('f2', 'String')])
 
         self.database.migrate('tests.sample_migrations', 12)
         self.assertEqual(self.database.count(Model3), 3)
@@ -88,29 +91,29 @@ class MigrationsTestCase(unittest.TestCase):
         self.assertListEqual(data, [1, 2, 3, 4])
 
         self.database.migrate('tests.sample_migrations', 14)
-        self.assertTrue(self.tableExists(MaterializedModel1))
-        self.assertEqual(self.getTableFields(MaterializedModel1),
+        self.assertTrue(self.table_exists(MaterializedModel1))
+        self.assertEqual(self.get_table_fields(MaterializedModel1),
                           [('date_time', 'DateTime'), ('int_field', 'Int8'), ('date', 'Date'), ('int_field_plus_one', 'Int8')])
-        self.assertTrue(self.tableExists(AliasModel1))
-        self.assertEqual(self.getTableFields(AliasModel1),
+        self.assertTrue(self.table_exists(AliasModel1))
+        self.assertEqual(self.get_table_fields(AliasModel1),
                           [('date', 'Date'), ('int_field', 'Int8'), ('date_alias', 'Date'), ('int_field_plus_one', 'Int8')])
         # Codecs and low cardinality
         self.database.migrate('tests.sample_migrations', 15)
-        self.assertTrue(self.tableExists(Model4_compressed))
+        self.assertTrue(self.table_exists(Model4_compressed))
         if self.database.has_low_cardinality_support:
-            self.assertEqual(self.getTableFields(Model2LowCardinality),
+            self.assertEqual(self.get_table_fields(Model2LowCardinality),
                              [('date', 'Date'), ('f1', 'LowCardinality(Int32)'), ('f3', 'LowCardinality(Float32)'),
                               ('f2', 'LowCardinality(String)'), ('f4', 'LowCardinality(Nullable(String))'), ('f5', 'Array(LowCardinality(UInt64))')])
         else:
             logging.warning('No support for low cardinality')
-            self.assertEqual(self.getTableFields(Model2),
+            self.assertEqual(self.get_table_fields(Model2),
                              [('date', 'Date'), ('f1', 'Int32'), ('f3', 'Float32'), ('f2', 'String'), ('f4', 'Nullable(String)'),
                               ('f5', 'Array(UInt64)')])
 
         if self.database.server_version >= (19, 14, 3, 3):
-            # Adding constraints
+            # Creating constraints
             self.database.migrate('tests.sample_migrations', 16)
-            self.assertTrue(self.tableExists(ModelWithConstraints))
+            self.assertTrue(self.table_exists(ModelWithConstraints))
             self.database.insert([ModelWithConstraints(f1=101, f2='a')])
             with self.assertRaises(ServerError):
                 self.database.insert([ModelWithConstraints(f1=99, f2='a')])
@@ -124,6 +127,19 @@ class MigrationsTestCase(unittest.TestCase):
             with self.assertRaises(ServerError):
                 self.database.insert([ModelWithConstraints(f1=99, f2='x')])
 
+        if self.database.server_version < (19, 3, 3):
+            # Creating indexes
+            self.database.migrate('tests.sample_migrations', 18)
+            self.assertTrue(self.table_exists(ModelWithIndex))
+            self.assertIn('INDEX `index`', self.get_table_def())
+            self.assertIn('INDEX another_index', self.get_table_def())
+            # Modifying indexes
+            self.database.migrate('tests.sample_migrations', 19)
+            self.assertNotIn('INDEX `index`', self.get_table_def())
+            self.assertIn('INDEX index2', self.get_table_def())
+            self.assertIn('INDEX another_index', self.get_table_def())
+
+
 # Several different models with the same table name, to simulate a table that changes over time
 
 class Model1(Model):
@@ -344,3 +360,35 @@ class ModelWithConstraints2(Model):
     def table_name(cls):
         return 'modelwithconstraints'
 
+
+class ModelWithIndex(Model):
+
+    date = DateField()
+    f1 = Int32Field()
+    f2 = StringField()
+
+    index = Index(f1, type=Index.minmax(), granularity=1)
+    another_index = Index(f2, type=Index.set(0), granularity=1)
+
+    engine = MergeTree('date', ('date',))
+
+    @classmethod
+    def table_name(cls):
+        return 'modelwithindex'
+
+
+class ModelWithIndex2(Model):
+
+    date = DateField()
+    f1 = Int32Field()
+    f2 = StringField()
+
+    index2 = Index(f1, type=Index.bloom_filter(), granularity=2)
+    another_index = Index(f2, type=Index.set(0), granularity=1)
+
+    engine = MergeTree('date', ('date',))
+
+    @classmethod
+    def table_name(cls):
+        return 'modelwithindex'
+