diff --git a/python/paddle/fluid/contrib/int8_inference/README.md b/python/paddle/fluid/contrib/int8_inference/README.md
index a9691dad4494f5eacf427b2806b2393baa57dc1e..460ae393f158ae320c93601365a68b8cfe2ba50e 100644
--- a/python/paddle/fluid/contrib/int8_inference/README.md
+++ b/python/paddle/fluid/contrib/int8_inference/README.md
@@ -63,10 +63,10 @@ Notes:
 ## 4. How to reproduce the results
 * Small dataset
 ```bash
-python python/paddle/fluid/contrib/tests/test_calibration.py
+FLAGS_use_mkldnn=true python python/paddle/fluid/contrib/tests/test_calibration.py
 ```
 
 * Full dataset
 ```bash
-DATASET=full python python/paddle/fluid/contrib/tests/test_calibration.py
+FLAGS_use_mkldnn=true DATASET=full python python/paddle/fluid/contrib/tests/test_calibration.py
 ```
diff --git a/python/paddle/fluid/contrib/tests/CMakeLists.txt b/python/paddle/fluid/contrib/tests/CMakeLists.txt
index 81aee1233d1db756686d1a934b94672dc5c770fe..a2c59416467e5dbe66f058666633807eb0e45047 100644
--- a/python/paddle/fluid/contrib/tests/CMakeLists.txt
+++ b/python/paddle/fluid/contrib/tests/CMakeLists.txt
@@ -6,5 +6,9 @@ if(APPLE OR WIN32 OR NOT WITH_MKL)
 endif()
 
 foreach(src ${TEST_OPS})
-    py_test(${src} SRCS ${src}.py)
+    if(src MATCHES "test_calibration")
+        py_test(${src} SRCS ${src}.py ENVS FLAGS_use_mkldnn=true)
+    else()
+        py_test(${src} SRCS ${src}.py)
+    endif()
 endforeach()
diff --git a/python/paddle/fluid/contrib/tests/test_calibration.py b/python/paddle/fluid/contrib/tests/test_calibration.py
index 424ea245a0f2dff0d437ace386f2e4e0fa6b517d..b9f938bebed71dc9611df8d743a066858ea38bca 100644
--- a/python/paddle/fluid/contrib/tests/test_calibration.py
+++ b/python/paddle/fluid/contrib/tests/test_calibration.py
@@ -199,7 +199,6 @@ class TestCalibrationForResnet50(unittest.TestCase):
 
     def run_program(self, model_path, generate_int8=False, algo='direct'):
         image_shape = [3, 224, 224]
-        os.environ['FLAGS_use_mkldnn'] = 'True'
 
         fluid.memory_optimize(fluid.default_main_program())
 
@@ -241,9 +240,6 @@ class TestCalibrationForResnet50(unittest.TestCase):
             label = label.reshape([-1, 1])
             running_program = calibrator.sampling_program.clone(
             ) if generate_int8 else infer_program.clone()
-            for op in running_program.current_block().ops:
-                if op.has_attr("use_mkldnn"):
-                    op._set_attr("use_mkldnn", True)
 
             t1 = time.time()
             _, acc1, _ = exe.run(