diff --git a/tensorflow/python/ops/numpy_ops/np_math_ops.py b/tensorflow/python/ops/numpy_ops/np_math_ops.py
index df3d7cf32ab..f9d3b34e90d 100644
--- a/tensorflow/python/ops/numpy_ops/np_math_ops.py
+++ b/tensorflow/python/ops/numpy_ops/np_math_ops.py
@@ -34,6 +34,7 @@ from tensorflow.python.ops import clip_ops
 from tensorflow.python.ops import control_flow_ops
 from tensorflow.python.ops import math_ops
 from tensorflow.python.ops import sort_ops
+from tensorflow.python.ops import special_math_ops
 from tensorflow.python.ops.numpy_ops import np_array_ops
 from tensorflow.python.ops.numpy_ops import np_arrays
 from tensorflow.python.ops.numpy_ops import np_dtypes
@@ -1336,3 +1337,30 @@ def meshgrid(*xi, **kwargs):
   outputs = [np_utils.tensor_to_ndarray(output) for output in outputs]
 
   return outputs
+
+
+@np_utils.np_doc('einsum')
+def einsum(subscripts, *operands, casting='safe', optimize=False):  # pylint: disable=missing-docstring
+  if casting == 'safe':
+    operands = np_array_ops._promote_dtype(*operands)  # pylint: disable=protected-access
+  elif casting == 'no':
+    operands = [np_array_ops.asarray(x) for x in operands]
+  else:
+    raise ValueError('casting policy not supported: %s' % casting)
+  if not optimize:
+    # TF doesn't have a "no optimization" option.
+    # TODO(wangpeng): Print a warning that np and tf use different
+    #   optimizations.
+    tf_optimize = 'greedy'
+  elif optimize == True:  # pylint: disable=singleton-comparison,g-explicit-bool-comparison
+    tf_optimize = 'greedy'
+  elif optimize == 'greedy':
+    tf_optimize = 'greedy'
+  elif optimize == 'optimal':
+    tf_optimize = 'optimal'
+  else:
+    raise ValueError('`optimize` method not supported: %s' % optimize)
+  operands = [x.data for x in operands]
+  res = special_math_ops.einsum(subscripts, *operands, optimize=tf_optimize)
+  res = np_utils.tensor_to_ndarray(res)
+  return res