diff --git a/spark/dl/src/main/scala/com/intel/analytics/bigdl/nn/mkldnn/Dropout.scala b/spark/dl/src/main/scala/com/intel/analytics/bigdl/nn/mkldnn/Dropout.scala
index 7e881efd24f..67f178e2bb3 100644
--- a/spark/dl/src/main/scala/com/intel/analytics/bigdl/nn/mkldnn/Dropout.scala
+++ b/spark/dl/src/main/scala/com/intel/analytics/bigdl/nn/mkldnn/Dropout.scala
@@ -49,6 +49,7 @@ class Dropout(
     _gradOutputFormats = grad.map(x => HeapData(x.shape, format(x.shape)))
     _gradOutputFormatsForWeight = grad.map(x => HeapData(x.shape, format(x.shape)))
     _gradInputFormats = grad.map(x => HeapData(x.shape, format(x.shape)))
+    _gradInputFormats.map(_.getPrimitive(runtime))
     gradInput = initTensor(_gradInputFormats.head)
     (_gradOutputFormats, _gradInputFormats)
   }
diff --git a/spark/dl/src/test/scala/com/intel/analytics/bigdl/nn/mkldnn/DropoutSpec.scala b/spark/dl/src/test/scala/com/intel/analytics/bigdl/nn/mkldnn/DropoutSpec.scala
index 5a45c236e7e..8fdf58b3e3e 100644
--- a/spark/dl/src/test/scala/com/intel/analytics/bigdl/nn/mkldnn/DropoutSpec.scala
+++ b/spark/dl/src/test/scala/com/intel/analytics/bigdl/nn/mkldnn/DropoutSpec.scala
@@ -67,4 +67,18 @@ class DropoutSpec extends FlatSpec with Matchers {
     val ratio = notEqZeros.toDouble / total
     ratio should be (1.0)
   }
+
+  "dropout in sequential" should "work correctly" in {
+    val shape = Array(2, 3, 4, 4)
+    val dropout = Dropout()
+    val seq = Sequential().add(Input(shape, Memory.Format.nchw))
+      .add(dropout)
+      .add(Output(Memory.Format.nchw))
+
+    seq.compile(TrainingPhase)
+
+    val input = Tensor[Float](shape).rand(-1, 1)
+    seq.forward(input)
+    seq.backward(input, seq.output)
+  }
 }