diff --git a/paddle/fluid/operators/cast_op.cc b/paddle/fluid/operators/cast_op.cc
index 72f8cb04f2de3af4ee526c3d9b86ff96e34f0b0a..dd0068d571f72c9c22334e523cd091fe4c8da5a6 100644
--- a/paddle/fluid/operators/cast_op.cc
+++ b/paddle/fluid/operators/cast_op.cc
@@ -14,6 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/cast_op.h"
 #include "paddle/fluid/framework/op_registry.h"
+#include "paddle/fluid/platform/float16.h"
 
 namespace paddle {
 namespace operators {
@@ -88,4 +89,5 @@ REGISTER_OP_CPU_KERNEL(cast, ops::CastOpKernel<CPU, float>,
                        ops::CastOpKernel<CPU, double>,
                        ops::CastOpKernel<CPU, int>,
                        ops::CastOpKernel<CPU, int64_t>,
-                       ops::CastOpKernel<CPU, bool>);
+                       ops::CastOpKernel<CPU, bool>,
+                       ops::CastOpKernel<CPU, paddle::platform::float16>);
diff --git a/paddle/fluid/operators/cast_op.cu b/paddle/fluid/operators/cast_op.cu
index 507e9a531aae70e60bc6748bfab800310d6e0c21..c486c5850e25fcf4370f02cb145c244743a4cc4b 100644
--- a/paddle/fluid/operators/cast_op.cu
+++ b/paddle/fluid/operators/cast_op.cu
@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/cast_op.h"
+#include "paddle/fluid/platform/float16.h"
 
 template <typename T>
 using CastOpKernel =
@@ -20,4 +21,5 @@ using CastOpKernel =
 
 REGISTER_OP_CUDA_KERNEL(cast, CastOpKernel<float>, CastOpKernel<double>,
                         CastOpKernel<int>, CastOpKernel<int64_t>,
-                        CastOpKernel<bool>);
+                        CastOpKernel<bool>,
+                        CastOpKernel<paddle::platform::float16>);
diff --git a/python/paddle/fluid/tests/unittests/test_cast_op.py b/python/paddle/fluid/tests/unittests/test_cast_op.py
index 8fb8d03828393ccfe57c0848d79b960c641ad39a..9c6ab00c85f8a6f735d1fab847c86f03664ee0d6 100644
--- a/python/paddle/fluid/tests/unittests/test_cast_op.py
+++ b/python/paddle/fluid/tests/unittests/test_cast_op.py
@@ -18,7 +18,7 @@ import numpy as np
 import paddle.fluid.core as core
 
 
-class TestCastOp(op_test.OpTest):
+class TestCastOp1(op_test.OpTest):
     def setUp(self):
         ipt = np.random.random(size=[10, 10])
         self.inputs = {'X': ipt.astype('float32')}
@@ -36,5 +36,36 @@ class TestCastOp(op_test.OpTest):
         self.check_grad(['X'], ['Out'])
 
 
+class TestCastOp2(op_test.OpTest):
+    def setUp(self):
+        ipt = np.random.random(size=[10, 10])
+        # numpy float16 is binded to fluid float16 via uint16
+        self.inputs = {'X': ipt.astype('float16').view(uint16)}
+        self.outputs = {'Out': ipt.astype('float32')}
+        self.attrs = {
+            'in_dtype': int(core.VarDesc.VarType.FP16),
+            'out_dtype': int(core.VarDesc.VarType.FP32)
+        }
+        self.op_type = 'cast'
+
+    def test_check_output(self):
+        self.check_output()
+
+
+class TestCastOp2(op_test.OpTest):
+    def setUp(self):
+        ipt = np.random.random(size=[10, 10])
+        self.inputs = {'X': ipt.astype('float32')}
+        self.outputs = {'Out': ipt.astype('float16')}
+        self.attrs = {
+            'in_dtype': int(core.VarDesc.VarType.FP32),
+            'out_dtype': int(core.VarDesc.VarType.FP16)
+        }
+        self.op_type = 'cast'
+
+    def test_check_output(self):
+        self.check_output()
+
+
 if __name__ == '__main__':
     unittest.main()