diff --git a/python/paddle/fluid/tests/unittests/test_norm_op.py b/python/paddle/fluid/tests/unittests/test_norm_op.py
index 00ea930251b6d0dcd025dc1ff0a549a6e403c734..15e033dc211e9ad8aa4dcac7d46e4010aff0afb3 100644
--- a/python/paddle/fluid/tests/unittests/test_norm_op.py
+++ b/python/paddle/fluid/tests/unittests/test_norm_op.py
@@ -16,7 +16,7 @@ from __future__ import print_function
 
 import unittest
 import numpy as np
-from op_test import OpTest
+from op_test import OpTest, skip_check_grad_ci
 
 
 def l2_norm(x, axis, epsilon):
@@ -63,6 +63,8 @@ class TestNormOp3(TestNormOp):
         self.epsilon = 1e-8
 
 
+@skip_check_grad_ci(reason="'check_grad' on large inputs is too slow, " +
+                    "however it is desirable to cover the forward pass")
 class TestNormOp4(TestNormOp):
     def init_test_case(self):
         self.shape = [128, 1024, 14, 14]
@@ -70,10 +72,11 @@ class TestNormOp4(TestNormOp):
         self.epsilon = 1e-8
 
     def test_check_grad(self):
-        # since the gradient check is very slow in large shape, so skip check_grad
         pass
 
 
+@skip_check_grad_ci(reason="'check_grad' on large inputs is too slow, " +
+                    "however it is desirable to cover the forward pass")
 class TestNormOp5(TestNormOp):
     def init_test_case(self):
         self.shape = [2048, 2048]
@@ -81,7 +84,6 @@ class TestNormOp5(TestNormOp):
         self.epsilon = 1e-8
 
     def test_check_grad(self):
-        # since the gradient check is very slow in large shape, so skip check_grad
         pass