diff --git a/python/paddle/fluid/layers/nn.py b/python/paddle/fluid/layers/nn.py
index 37ce73827532d4fbaeffc5dc798b0149de371be1..56c37f05ccd6d0f228c96a43cade71ebbe013f7e 100644
--- a/python/paddle/fluid/layers/nn.py
+++ b/python/paddle/fluid/layers/nn.py
@@ -1518,9 +1518,6 @@ def batch_norm(input,
     bias = helper.create_parameter(
         attr=helper.bias_attr, shape=param_shape, dtype=dtype, is_bias=True)
 
-    if do_model_average_for_mean_and_var:
-        do_model_average_for_mean_and_var = None
-
     mean = helper.create_parameter(
         attr=ParamAttr(
             name=moving_mean_name,
diff --git a/python/paddle/fluid/optimizer.py b/python/paddle/fluid/optimizer.py
index 1917b7d044c565d351f0a9de72437571064221b2..36503cac6d5391821b977d90e6b77c4df7e3b564 100644
--- a/python/paddle/fluid/optimizer.py
+++ b/python/paddle/fluid/optimizer.py
@@ -853,7 +853,8 @@ class ModelAverage(Optimizer):
         self.params_grads = [] if params_grads is None else params_grads
         params = {}
         for param, grad in self.params_grads:
-            params[param.name] = (param, grad)
+            if param.do_model_average != False:
+                params[param.name] = (param, grad)
         for param in framework.default_main_program().global_block(
         ).all_parameters():
             if param.name not in params and param.do_model_average != False: