diff --git a/paddle/fluid/operators/mkldnn/activation_mkldnn_op.cc b/paddle/fluid/operators/mkldnn/activation_mkldnn_op.cc
index 86fe40c4f6a825116cdf8fe884ae06cc3e7bbc34..aa9606b5f85896cf4905c53b655f894e6429fc9a 100644
--- a/paddle/fluid/operators/mkldnn/activation_mkldnn_op.cc
+++ b/paddle/fluid/operators/mkldnn/activation_mkldnn_op.cc
@@ -196,6 +196,10 @@ template <typename T>
 using SwishMKLDNNFunctor =
     MKLDNNActivationFunc<T, mkldnn::algorithm::eltwise_swish>;
 
+template <typename T>
+using SigmoidMKLDNNFunctor =
+    MKLDNNActivationFunc<T, mkldnn::algorithm::eltwise_logistic>;
+
 template <typename T>
 using TanhMKLDNNFunctor =
     MKLDNNActivationFunc<T, mkldnn::algorithm::eltwise_tanh>;
@@ -216,6 +220,10 @@ template <typename T>
 using SwishMKLDNNGradFunctor =
     MKLDNNActivationGradFunc<T, mkldnn::algorithm::eltwise_swish>;
 
+template <typename T>
+using SigmoidMKLDNNGradFunctor =
+    MKLDNNActivationGradFunc<T, mkldnn::algorithm::eltwise_logistic>;
+
 template <typename T>
 using TanhMKLDNNGradFunctor =
     MKLDNNActivationGradFunc<T, mkldnn::algorithm::eltwise_tanh>;
@@ -239,13 +247,14 @@ namespace ops = paddle::operators;
       act_type##_grad, MKLDNN, ::paddle::platform::CPUPlace,               \
       ops::MKLDNNActivationGradKernel<ops::grad_functor<float>>);
 
-#define FOR_EACH_MKLDNN_KERNEL_FUNCTOR(__macro)                  \
-  __macro(relu, ReluMKLDNNFunctor, ReluMKLDNNGradFunctor);       \
-  __macro(leaky_relu, ReluMKLDNNFunctor, ReluMKLDNNGradFunctor); \
-  __macro(gelu, GeluMKLDNNFunctor, GeluMKLDNNGradFunctor);       \
-  __macro(swish, SwishMKLDNNFunctor, SwishMKLDNNGradFunctor);    \
-  __macro(tanh, TanhMKLDNNFunctor, TanhMKLDNNGradFunctor);       \
-  __macro(sqrt, SqrtMKLDNNFunctor, SqrtMKLDNNGradFunctor);       \
+#define FOR_EACH_MKLDNN_KERNEL_FUNCTOR(__macro)                     \
+  __macro(relu, ReluMKLDNNFunctor, ReluMKLDNNGradFunctor);          \
+  __macro(leaky_relu, ReluMKLDNNFunctor, ReluMKLDNNGradFunctor);    \
+  __macro(gelu, GeluMKLDNNFunctor, GeluMKLDNNGradFunctor);          \
+  __macro(swish, SwishMKLDNNFunctor, SwishMKLDNNGradFunctor);       \
+  __macro(sigmoid, SigmoidMKLDNNFunctor, SigmoidMKLDNNGradFunctor); \
+  __macro(tanh, TanhMKLDNNFunctor, TanhMKLDNNGradFunctor);          \
+  __macro(sqrt, SqrtMKLDNNFunctor, SqrtMKLDNNGradFunctor);          \
   __macro(abs, AbsMKLDNNFunctor, AbsMKLDNNGradFunctor);
 
 FOR_EACH_MKLDNN_KERNEL_FUNCTOR(REGISTER_ACTIVATION_MKLDNN_KERNEL);
diff --git a/python/paddle/fluid/tests/unittests/mkldnn/test_activation_mkldnn_op.py b/python/paddle/fluid/tests/unittests/mkldnn/test_activation_mkldnn_op.py
index 9f635c3f267625311807c3cf3520957d08ade3c3..2404aeb72b2a77f1f817cec697b3188003e884eb 100644
--- a/python/paddle/fluid/tests/unittests/mkldnn/test_activation_mkldnn_op.py
+++ b/python/paddle/fluid/tests/unittests/mkldnn/test_activation_mkldnn_op.py
@@ -19,7 +19,7 @@ import numpy as np
 from scipy.special import expit
 import paddle.fluid.core as core
 from paddle.fluid.tests.unittests.op_test import OpTest
-from paddle.fluid.tests.unittests.test_activation_op import TestActivation, TestRelu, TestTanh, TestSqrt, TestAbs, TestLeakyRelu, TestSwish
+from paddle.fluid.tests.unittests.test_activation_op import TestActivation, TestRelu, TestTanh, TestSqrt, TestAbs, TestLeakyRelu, TestSwish, TestSigmoid
 from paddle.fluid.tests.unittests.test_gelu_op import gelu
 from mkldnn_op_test import check_if_mkldnn_primitives_exist_in_bwd
 
@@ -162,6 +162,12 @@ class TestMKLDNNSwishDim2(TestSwish):
         self.check_grad(['X'], 'Out')
 
 
+class TestMKLDNNSigmoidDim2(TestSigmoid):
+    def setUp(self):
+        super(TestMKLDNNSigmoidDim2, self).setUp()
+        self.attrs = {"use_mkldnn": True}
+
+
 class TestMKLDNNReluDim4(TestRelu):
     def setUp(self):
         super(TestMKLDNNReluDim4, self).setUp()
@@ -328,6 +334,17 @@ class TestMKLDNNSwishDim4(TestSwish):
         self.check_grad(['X'], 'Out')
 
 
+class TestMKLDNNSigmoidDim4(TestSigmoid):
+    def setUp(self):
+        super(TestMKLDNNSigmoidDim4, self).setUp()
+
+        x = np.random.uniform(0.1, 1, [2, 4, 3, 5]).astype(self.dtype)
+        out = 1 / (1 + np.exp(-x))
+        self.inputs = {'X': OpTest.np_dtype_to_fluid_dtype(x)}
+        self.outputs = {'Out': out}
+        self.attrs = {"use_mkldnn": True}
+
+
 # Check if primitives already exist in backward
 class TestMKLDNNAbsPrimitivesAlreadyExist(unittest.TestCase):
     def setUp(self):
diff --git a/python/paddle/fluid/tests/unittests/test_imperative_basic.py b/python/paddle/fluid/tests/unittests/test_imperative_basic.py
index 75661644c1b41e22e4b3c5de9dd32760aaf99191..9b6c307bbec5d272aa3c5644aeaabfe9d7f5df8f 100644
--- a/python/paddle/fluid/tests/unittests/test_imperative_basic.py
+++ b/python/paddle/fluid/tests/unittests/test_imperative_basic.py
@@ -626,7 +626,7 @@ class TestDygraphUtils(unittest.TestCase):
             a = fluid.dygraph.to_variable(a_np)
             res1 = func(a, act="sigmoid", use_mkldnn=True, use_cudnn=True)
             res2 = fluid.layers.sigmoid(a)
-            self.assertTrue(np.array_equal(res1.numpy(), res2.numpy()))
+            self.assertTrue(np.allclose(res1.numpy(), res2.numpy()))
 
     def test_append_bias_in_dygraph_exception(self):
         with new_program_scope():