diff --git a/PaddleCV/PaddleGAN/network/CGAN_network.py b/PaddleCV/PaddleGAN/network/CGAN_network.py
index 4978f6dd8e57c3a0f2a6c0fad06fcb2770a88efe..8b65b7bd32b21ed78c222658d151117b692ebc10 100644
--- a/PaddleCV/PaddleGAN/network/CGAN_network.py
+++ b/PaddleCV/PaddleGAN/network/CGAN_network.py
@@ -35,6 +35,10 @@ class CGAN_model(object):
         self.gf_dim = 128
         self.df_dim = 64
         self.leaky_relu_factor = 0.2
+        if self.batch_size == 1:
+            self.norm = None
+        else:
+            self.norm = "batch_norm"
 
     def network_G(self, input, label, name="generator"):
         # concat noise and label
@@ -43,14 +47,14 @@ class CGAN_model(object):
         o_l1 = linear(
             xy,
             self.gf_dim * 8,
-            norm='batch_norm',
+            norm=self.norm,
             activation_fn='relu',
             name=name + '_l1')
         o_c1 = fluid.layers.concat([o_l1, y], 1)
         o_l2 = linear(
             o_c1,
             self.gf_dim * (self.img_w // 4) * (self.img_h // 4),
-            norm='batch_norm',
+            norm=self.norm,
             activation_fn='relu',
             name=name + '_l2')
         o_r1 = fluid.layers.reshape(
@@ -107,7 +111,7 @@ class CGAN_model(object):
         o_l3 = linear(
             o_c2,
             self.df_dim * 16,
-            norm='batch_norm',
+            norm=self.norm,
             activation_fn='leaky_relu',
             name=name + '_l3')
         o_c3 = fluid.layers.concat([o_l3, y], 1)
diff --git a/PaddleCV/PaddleGAN/network/DCGAN_network.py b/PaddleCV/PaddleGAN/network/DCGAN_network.py
index 6c01fa449e91d338c0b24de1301c93462e9a26d4..1b455c97a40781e4d0b82f192b34a8b7e6612a21 100644
--- a/PaddleCV/PaddleGAN/network/DCGAN_network.py
+++ b/PaddleCV/PaddleGAN/network/DCGAN_network.py
@@ -31,13 +31,17 @@ class DCGAN_model(object):
         self.dfc_dim = 1024
         self.gf_dim = 64
         self.df_dim = 64
+        if self.batch_size == 1:
+            self.norm = None
+        else:
+            self.norm = "batch_norm"
 
     def network_G(self, input, name="generator"):
-        o_l1 = linear(input, self.gfc_dim, norm='batch_norm', name=name + '_l1')
+        o_l1 = linear(input, self.gfc_dim, norm=self.norm, name=name + '_l1')
         o_l2 = linear(
             o_l1,
             self.gf_dim * 2 * self.img_dim // 4 * self.img_dim // 4,
-            norm='batch_norm',
+            norm=self.norm,
             name=name + '_l2')
         o_r1 = fluid.layers.reshape(
             o_l2, [-1, self.df_dim * 2, self.img_dim // 4, self.img_dim // 4])
@@ -85,7 +89,7 @@ class DCGAN_model(object):
         o_l1 = linear(
             o_c2,
             self.dfc_dim,
-            norm='batch_norm',
+            norm=self.norm,
             activation_fn='leaky_relu',
             name=name + '_l1')
         out = linear(o_l1, 1, activation_fn='sigmoid', name=name + '_l2')