diff --git a/paddle/fluid/operators/conv_op.cc b/paddle/fluid/operators/conv_op.cc
index 4cd20e6ecd5e6..1610705c4694c 100644
--- a/paddle/fluid/operators/conv_op.cc
+++ b/paddle/fluid/operators/conv_op.cc
@@ -50,6 +50,15 @@ std::vector<int64_t> ConvOp::ComputeOutputShape(
       ctx->Attrs().Get<std::string>("padding_algorithm");
   int groups = ctx->Attrs().Get<int>("groups");
   std::vector<int> dilations = ctx->Attrs().Get<std::vector<int>>("dilations");
+  int dilation_size = dilations.size();
+  for (int i = 0; i < dilation_size; ++i) {
+    PADDLE_ENFORCE_GT(
+        dilations[i], 0,
+        platform::errors::InvalidArgument(
+            "The dilation of Op(Conv) should be larget than 0, but received "
+            "dilation is %d.",
+            dilations[i]));
+  }
   const std::string data_format = ctx->Attrs().Get<std::string>("data_format");
 
   // MKL-DNN Kernels are using NCHW order of dims description
diff --git a/python/paddle/fluid/tests/unittests/test_conv1d_layer.py b/python/paddle/fluid/tests/unittests/test_conv1d_layer.py
index fc0a64b18a7af..715579c332dfa 100644
--- a/python/paddle/fluid/tests/unittests/test_conv1d_layer.py
+++ b/python/paddle/fluid/tests/unittests/test_conv1d_layer.py
@@ -204,6 +204,7 @@ def add_error_cases(suite):
     suite.addTest(
         Conv1DErrorTestCase(
             methodName='runTest', padding=[1, 2, 3, 4, 5]))
+    suite.addTest(Conv1DErrorTestCase(methodName='runTest', dilation=-10))
 
 
 def load_tests(loader, standard_tests, pattern):