diff --git a/dataset/table.py b/dataset/table.py
index b7bf815..e59dbb2 100644
--- a/dataset/table.py
+++ b/dataset/table.py
@@ -12,7 +12,7 @@ from sqlalchemy.exc import NoSuchTableError
 from dataset.types import Types
 from dataset.util import normalize_column_name, index_name, ensure_tuple
 from dataset.util import DatasetException, ResultIter, QUERY_STEP
-from dataset.util import normalize_table_name
+from dataset.util import normalize_table_name, pad_chunk_columns
 
 
 log = logging.getLogger(__name__)
@@ -127,10 +127,12 @@ class Table(object):
             row = self._sync_columns(row, ensure, types=types)
             chunk.append(row)
             if len(chunk) == chunk_size:
+                chunk = pad_chunk_columns(chunk)
                 self.table.insert().execute(chunk)
                 chunk = []
 
         if len(chunk):
+            chunk = pad_chunk_columns(chunk)
             self.table.insert().execute(chunk)
 
     def update(self, row, keys, ensure=None, types=None, return_count=False):
diff --git a/dataset/util.py b/dataset/util.py
index 2a82d15..04313cd 100644
--- a/dataset/util.py
+++ b/dataset/util.py
@@ -102,3 +102,15 @@ def ensure_tuple(obj):
     if isinstance(obj, Sequence) and not isinstance(obj, six.string_types):
         return tuple(obj)
     return obj,
+
+
+def pad_chunk_columns(chunk):
+    """Given a set of items to be inserted, make sure they all have the
+    same columns by padding columns with None if they are missing."""
+    columns = set()
+    for record in chunk:
+        columns.update(record.keys())
+    for record in chunk:
+        for column in columns:
+            record.setdefault(column, None)
+    return chunk
\ No newline at end of file