diff --git a/demo/mnist/api_train_v2.py b/demo/mnist/api_train_v2.py
index a23ddfaca011e2057af15fc4c559080c29eecc73..8a612cbc66d54ba3a3a9a444f85626812940cf6a 100644
--- a/demo/mnist/api_train_v2.py
+++ b/demo/mnist/api_train_v2.py
@@ -41,7 +41,7 @@ def main():
 
     trainer.train(
         train_data_reader=train_reader,
-        topology=cost,
+        cost=cost,
         parameters=parameters,
         event_handler=event_handler,
         batch_size=32,  # batch size should be refactor in Data reader
diff --git a/python/paddle/v2/trainer.py b/python/paddle/v2/trainer.py
index 3bf2128e16b5ad17968bd4c8debf577b86c50414..be33b910807ff30eb8ea9b37d5b172a556a1c3f5 100644
--- a/python/paddle/v2/trainer.py
+++ b/python/paddle/v2/trainer.py
@@ -62,7 +62,7 @@ class SGD(ITrainer):
 
     def train(self,
               train_data_reader,
-              topology,
+              cost,
               parameters,
               num_passes=1,
               test_data_reader=None,
@@ -73,7 +73,7 @@ class SGD(ITrainer):
         Training method. Will train num_passes of input data.
 
         :param train_data_reader:
-        :param topology: cost layers, use one or more Layers to represent it.
+        :param cost: cost layers, to be optimized.
         :param parameters: The parameter pools.
         :param num_passes: The total train passes.
         :param test_data_reader:
@@ -86,7 +86,7 @@ class SGD(ITrainer):
         if event_handler is None:
             event_handler = default_event_handler
 
-        topology = Topology(topology)
+        topology = Topology(cost)
 
         __check_train_args__(**locals())