diff --git a/nnvm/src/top/tensor/broadcast.cc b/nnvm/src/top/tensor/broadcast.cc
index edf209e3e9ec..a3ac2df93f56 100644
--- a/nnvm/src/top/tensor/broadcast.cc
+++ b/nnvm/src/top/tensor/broadcast.cc
@@ -337,8 +337,7 @@ Example::
       });
     NodeEntry dy = MakeNode("broadcast_div", n->attrs.name + "_drhs_div", {
         NodeEntry{n, 0, 0},
-        MakeNode("__mul_scalar__", n->attrs.name + "_rhs_by_two",
-                 {n->inputs[1]}, {{"scalar", "2"}})
+        MakeNode("negative", n->attrs.name + "_rhs_neg", {n->inputs[1]})
       });
     NodeEntry drhs = MakeNode("collapse_sum", n->attrs.name + "_drhs_sum", {
         MakeNode("broadcast_mul", n->attrs.name + "_drhs_mul", { dy, ograds[0] }),
diff --git a/nnvm/tests/python/compiler/test_top_level4.py b/nnvm/tests/python/compiler/test_top_level4.py
index b7f759c62f99..5bf134b49a7b 100644
--- a/nnvm/tests/python/compiler/test_top_level4.py
+++ b/nnvm/tests/python/compiler/test_top_level4.py
@@ -268,7 +268,7 @@ def _backward_mul(head_grads, a, b):
     y = sym.broadcast_div(a, b)
     def _backward_div(head_grads, a, b):
         da = head_grads / b
-        db = _collapse(head_grads * a / (2 * b**2))
+        db = _collapse(- head_grads * a / b**2)
         return da, db
     helper(y, inputs, dtype, lambda a, b: a / b, _backward_div)