diff --git a/.gitignore b/.gitignore
index 9aa5b22..03e17bb 100644
--- a/.gitignore
+++ b/.gitignore
@@ -83,6 +83,11 @@ celerybeat-schedule
 venv/
 ENV/
 
+# Pipenv
+Pipfile
+Pipfile.lock
+pyproject.toml
+
 # Spyder project settings
 .spyderproject
 
diff --git a/singer/__init__.py b/singer/__init__.py
index e0deeeb..e9d33ab 100644
--- a/singer/__init__.py
+++ b/singer/__init__.py
@@ -26,6 +26,7 @@
     RecordMessage,
     SchemaMessage,
     StateMessage,
+    BatchMessage,
     format_message,
     parse_message,
     write_message,
@@ -34,6 +35,7 @@
     write_schema,
     write_state,
     write_version,
+    write_batch
 )
 
 from singer.transform import (
diff --git a/singer/messages.py b/singer/messages.py
index 3848801..61cc515 100644
--- a/singer/messages.py
+++ b/singer/messages.py
@@ -167,6 +167,52 @@ def asdict(self):
         }
 
 
+class BatchMessage(Message):
+    """ BATCH message (EXPERIMENTAL).
+
+    The BATCH message has these fields:
+
+      * stream (string) - The name of the stream.
+      * filepath (string) - The location of a batch file. e.g. '/tmp/users001.jsonl'.
+      * format (string, optional) - An indication of serialization format.
+            If none is provided, 'jsonl' will be assumed. e.g. 'csv'.
+      * compression (string, optional) - An indication of file compression format. e.g. 'gzip'.
+      * batch_size (int, optional) - Number of records in this batch. e.g. 100000.
+
+    If file_properties are not provided, uncompressed jsonl files are assumed.
+
+    A BATCH record points to a collection of messages (from a single stream) serialized to disk,
+    and is implemented for performance reasons. Most Taps and Targets should not need to use
+    BATCH messages at all.
+
+    msg = singer.BatchMessage(
+        stream='users',
+        filepath='/tmp/users0001.jsonl'
+    )
+
+    """
+
+    def __init__(self, stream, filepath, file_format=None, compression=None, batch_size=None):
+        self.stream = stream
+        self.filepath = filepath
+        self.format = file_format or 'jsonl'
+        self.compression = compression
+        self.batch_size = batch_size
+
+    def asdict(self):
+        result = {
+            'type': 'BATCH',
+            'stream': self.stream,
+            'filepath': self.filepath,
+            'format': self.format
+        }
+        if self.compression is not None:
+            result['compression'] = self.compression
+        if self.batch_size is not None:
+            result['batch_size'] = self.batch_size
+        return result
+
+
 def _required_key(msg, k):
     if k not in msg:
         raise Exception("Message is missing required key '{}': {}".format(k, msg))
@@ -214,6 +260,14 @@ def parse_message(msg):
     elif msg_type == 'ACTIVATE_VERSION':
         return ActivateVersionMessage(stream=_required_key(obj, 'stream'),
                                       version=_required_key(obj, 'version'))
+
+    elif msg_type == 'BATCH':
+        return BatchMessage(stream=_required_key(obj, 'stream'),
+                            filepath=_required_key(obj, 'filepath'),
+                            file_format=_required_key(obj, 'format'),
+                            compression=obj.get('compression'),
+                            batch_size=obj.get('batch_size'))
+
     else:
         return None
 
@@ -285,3 +339,17 @@ def write_version(stream_name, version):
     write_version(stream, version)
     """
     write_message(ActivateVersionMessage(stream_name, version))
+
+def write_batch(
+    stream_name, filepath, file_format=None,
+    compression=None, batch_size=None
+):
+    """Write a batch message.
+
+    stream = 'users'
+    filepath = '/tmp/users0001.jsonl'
+    file_format = 'jsonl'
+    compression = None
+    batch_size = 100000
+    """
+    write_message(BatchMessage(stream_name, filepath, file_format, compression, batch_size))
diff --git a/tests/test_singer.py b/tests/test_singer.py
index 4fb74de..8f24262 100644
--- a/tests/test_singer.py
+++ b/tests/test_singer.py
@@ -90,6 +90,15 @@ def test_parse_message_state_missing_value(self):
         with self.assertRaises(Exception):
             singer.parse_message('{"type": "STATE"}')
 
+    def test_parse_message_batch_good(self):
+        message = singer.parse_message(
+            '{"type": "BATCH", "stream": "users", "filepath": "/tmp/users0001.jsonl", "format": "jsonl"}')
+        self.assertEqual(message, singer.BatchMessage(stream='users', filepath='/tmp/users0001.jsonl'))
+
+    def test_parse_message_batch_missing_value(self):
+        with self.assertRaises(Exception):
+            singer.parse_message('{"type": "BATCH"}')
+
     def test_round_trip(self):
         record_message = singer.RecordMessage(
             record={'name': 'foo'},
@@ -112,7 +121,6 @@ def test_round_trip(self):
                          singer.parse_message(singer.format_message(state_message)))
 
     ## These three tests just confirm that writing doesn't throw
-
     def test_write_record(self):
         singer.write_record("users", {"name": "mike"})
 
@@ -125,6 +133,10 @@ def test_write_schema(self):
     def test_write_state(self):
         singer.write_state({"foo": 1})
 
+    def test_write_batch(self):
+        singer.write_batch("users", "/tmp/users0001.jsonl")
+
+
 class TestParsingNumbers(unittest.TestCase):
     def create_record(self, value):
         raw = '{"type": "RECORD", "stream": "test", "record": {"value": ' + value + '}}'