diff --git a/paddle/fluid/operators/concat_op.cc b/paddle/fluid/operators/concat_op.cc
index e52d280836e0236bea8ce871116da5d14f4d6bd6..c76ccb70f769644cf0996109c645b94223047510 100644
--- a/paddle/fluid/operators/concat_op.cc
+++ b/paddle/fluid/operators/concat_op.cc
@@ -41,8 +41,9 @@ class ConcatOp : public framework::OperatorWithKernel {
                     static_cast<int64_t>(ins[0].size()));
 
     const size_t n = ins.size();
-
-    PADDLE_ENFORCE_GT(n, 0, "Input tensors count should > 0.");
+    PADDLE_ENFORCE_GT(n, 0,
+                      "ShapeError: Input tensors count should > 0. But "
+                      "recevied inputs' length is 0.");
     if (n == 1) {
       VLOG(3) << "Warning: concat op have only one input, may waste memory";
     }
@@ -66,9 +67,14 @@ class ConcatOp : public framework::OperatorWithKernel {
               ctx->IsRuntime() || (out_dims[j] > 0 && ins[i][j] > 0);
           if (check_shape) {
             // check all shape in run time
-            PADDLE_ENFORCE_EQ(out_dims[j], ins[i][j],
-                              "Input tensors should have the same "
-                              "elements except the specify axis.");
+            PADDLE_ENFORCE_EQ(
+                out_dims[j], ins[i][j],
+                "ShapeError: Input tensors should have same "
+                "dimensions(or specific dimension = -1) except the axis. "
+                "But recevied axis = %s, input[0]'s shape = "
+                "[%s], input[%s]'s shape = [%s], the \"%s\" "
+                "dimension of input[%s] is unexpected",
+                axis, ins[0], i, ins[j], j, i);
           }
         }
       }
diff --git a/python/paddle/fluid/layers/tensor.py b/python/paddle/fluid/layers/tensor.py
index 4523dfd9589e83daf7c796415d0c224cfb39dc12..8f086952a4fa14b7b337756486f018d037fa9a5b 100644
--- a/python/paddle/fluid/layers/tensor.py
+++ b/python/paddle/fluid/layers/tensor.py
@@ -23,6 +23,8 @@ from ..core import VarDesc
 from .layer_function_generator import templatedoc
 from ..data_feeder import convert_dtype
 import numpy
+import warnings
+from ..data_feeder import convert_dtype
 
 __all__ = [
     'create_tensor', 'create_parameter', 'create_global_var', 'cast',
@@ -247,6 +249,21 @@ def concat(input, axis=0, name=None):
                 #  [14 15 16]]
     """
     helper = LayerHelper('concat', **locals())
+    for x in input:
+        if not isinstance(x, Variable):
+            raise TypeError(
+                "The type of x in 'input' in concat must be Variable, but received %s"
+                % (type(x)))
+        if convert_dtype(x.dtype) in ['float16']:
+            warnings.warn(
+                "The data type of x in 'input' in concat only support float16 on GPU now."
+            )
+        if convert_dtype(x.dtype) not in [
+                'float16', 'float32', 'float64', 'int32', 'int64'
+        ]:
+            raise TypeError(
+                "The data type of x in 'input' in concat must be float16(only support on GPU), float32, float64, int32, int64, but received %s."
+                % (convert_dtype(x.dtype)))
     out = helper.create_variable_for_type_inference(dtype=helper.input_dtype())
     helper.append_op(
         type='concat',
diff --git a/python/paddle/fluid/tests/unittests/test_concat_op.py b/python/paddle/fluid/tests/unittests/test_concat_op.py
index 3ba2002a66a979870708684f88415f65ca28362e..7055e4e2514f53530d3b9c843358a69fef9d7a7a 100644
--- a/python/paddle/fluid/tests/unittests/test_concat_op.py
+++ b/python/paddle/fluid/tests/unittests/test_concat_op.py
@@ -17,6 +17,8 @@ from __future__ import print_function
 import unittest
 import numpy as np
 from op_test import OpTest
+import paddle.fluid as fluid
+from paddle.fluid import compiler, Program, program_guard
 
 
 class TestConcatOp(OpTest):
@@ -112,5 +114,18 @@ create_test_fp16(TestConcatOp3)
 create_test_fp16(TestConcatOp4)
 create_test_fp16(TestConcatOp5)
 
+
+class TestConcatOpError(OpTest):
+    def test_errors(self):
+        with program_guard(Program(), Program()):
+            # The input type of concat_op must be Variable.
+            x1 = fluid.create_lod_tensor(
+                np.array([[-1]]), [[1]], fluid.CPUPlace())
+            self.assertRaises(TypeError, fluid.layers.concat, x1)
+            # The input dtype of concat_op must be float16(only support on GPU), float32, float64, int32, int64.
+            x2 = fluid.layers.data(name='x2', shape=[4], dtype='uint8')
+            self.assertRaises(TypeError, fluid.layers.concat, x2)
+
+
 if __name__ == '__main__':
     unittest.main()