diff --git a/spacy/cli/train.py b/spacy/cli/train.py
index a2c06c571..bc0664917 100644
--- a/spacy/cli/train.py
+++ b/spacy/cli/train.py
@@ -28,15 +28,17 @@ from .. import displacy
     n_iter=("number of iterations", "option", "n", int),
     n_sents=("number of sentences", "option", "ns", int),
     use_gpu=("Use GPU", "flag", "G", bool),
+    resume=("Whether to resume training", "flag", "R", bool),
     no_tagger=("Don't train tagger", "flag", "T", bool),
     no_parser=("Don't train parser", "flag", "P", bool),
     no_entities=("Don't train NER", "flag", "N", bool)
 )
 def train(cmd, lang, output_dir, train_data, dev_data, n_iter=20, n_sents=0,
-          use_gpu=False, no_tagger=False, no_parser=False, no_entities=False):
+          use_gpu=False, resume=False, no_tagger=False, no_parser=False, no_entities=False):
     """
     Train a model. Expects data in spaCy's JSON format.
     """
+    util.set_env_log(True)
     n_sents = n_sents or None
     output_path = util.ensure_path(output_dir)
     train_path = util.ensure_path(train_data)
@@ -66,7 +68,11 @@ def train(cmd, lang, output_dir, train_data, dev_data, n_iter=20, n_sents=0,
                                    util.env_opt('batch_to', 64),
                                    util.env_opt('batch_compound', 1.001))
 
-    nlp = lang_class(pipeline=pipeline)
+    if resume:
+        prints(output_path / 'model19.pickle', title="Resuming training")
+        nlp = dill.load((output_path / 'model19.pickle').open('rb'))
+    else:
+        nlp = lang_class(pipeline=pipeline)
     corpus = GoldCorpus(train_path, dev_path, limit=n_sents)
     n_train_docs = corpus.count_train()
 
@@ -75,6 +81,8 @@ def train(cmd, lang, output_dir, train_data, dev_data, n_iter=20, n_sents=0,
     print("Itn.\tLoss\tUAS\tNER P.\tNER R.\tNER F.\tTag %\tToken %")
     try:
         for i in range(n_iter):
+            if resume:
+                i += 20
             with tqdm.tqdm(total=corpus.count_train(), leave=False) as pbar:
                 train_docs = corpus.train_docs(nlp, projectivize=True,
                                                gold_preproc=False, max_length=0)
@@ -86,14 +94,18 @@ def train(cmd, lang, output_dir, train_data, dev_data, n_iter=20, n_sents=0,
                     pbar.update(len(docs))
 
             with nlp.use_params(optimizer.averages):
+                util.set_env_log(False)
+                epoch_model_path = output_path / ('model%d' % i)
+                nlp.to_disk(epoch_model_path)
                 with (output_path / ('model%d.pickle' % i)).open('wb') as file_:
                     dill.dump(nlp, file_, -1)
-                with (output_path / ('model%d.bin' % i)).open('wb') as file_:
-                    file_.write(nlp.to_bytes())
-                with (output_path / ('model%d.bin' % i)).open('rb') as file_:
-                    nlp_loaded = lang_class(pipeline=pipeline)
-                    nlp_loaded.from_bytes(file_.read())
-                scorer = nlp_loaded.evaluate(corpus.dev_docs(nlp_loaded, gold_preproc=False))
+                nlp_loaded = lang_class(pipeline=pipeline)
+                nlp_loaded = nlp_loaded.from_disk(epoch_model_path)
+                scorer = nlp_loaded.evaluate(
+                            corpus.dev_docs(
+                                nlp_loaded,
+                                gold_preproc=False))
+                util.set_env_log(True)
             print_progress(i, losses, scorer.scores)
     finally:
         print("Saving model...")