diff --git a/python/paddle/fluid/framework.py b/python/paddle/fluid/framework.py
index 96587b6e904f681a71182ffdb03608b5edde5e46..6f6d94a23dc194ac99aee171299b5ffd80b4e5f3 100644
--- a/python/paddle/fluid/framework.py
+++ b/python/paddle/fluid/framework.py
@@ -1725,6 +1725,18 @@ class Program(object):
         self._trainers_endpoints = []
         # the distributed lookup table names
         self._distributed_lookup_table = None
+        # whether the program is optimized by memory_optimize_transpiler
+        self.__is_optimized = False
+
+    @property
+    def _is_optimized(self):
+        # if the program is optimized, operator input/outputs
+        # maybe same, which conflict with save_inference_model.
+        return self.__is_optimized
+
+    @_is_optimized.setter
+    def set__is_optimized(self, target):
+        self.__is_optimized = target
 
     @property
     def op_role(self):
diff --git a/python/paddle/fluid/io.py b/python/paddle/fluid/io.py
index 6b1d4cc34f3cd40c878740f28618f26d5e89a6bd..836b28a5611a73e2f74248f7f1deb70b07a9ee6d 100644
--- a/python/paddle/fluid/io.py
+++ b/python/paddle/fluid/io.py
@@ -16,6 +16,7 @@ from __future__ import print_function
 
 import os
 import errno
+import warnings
 import time
 import shutil
 import six
@@ -930,6 +931,13 @@ def save_inference_model(dirname,
 
     if main_program is None:
         main_program = default_main_program()
+        if main_program.is_optimized:
+            warnings.warn(
+                "save_inference_model must put before you call memory_optimize. \
+                                            the memory_optimize will modify the original program, \
+                                            is not suitable for saving inference model \
+                                            we save the original program as inference model.",
+                RuntimeWarning)
 
     # when a pserver and a trainer running on the same machine, mkdir may conflict
     try:
diff --git a/python/paddle/fluid/transpiler/memory_optimization_transpiler.py b/python/paddle/fluid/transpiler/memory_optimization_transpiler.py
index e5d48d3d19ed71624d528144f13e23770a09362a..2e4dbfcdc956cbe42573b196c4a885f9038f9e31 100755
--- a/python/paddle/fluid/transpiler/memory_optimization_transpiler.py
+++ b/python/paddle/fluid/transpiler/memory_optimization_transpiler.py
@@ -540,6 +540,7 @@ def memory_optimize(input_program,
     if skip_opt_set is not None:
         skip_opt_set = set(map(to_name_str, skip_opt_set))
     cfgs = _get_cfgs(input_program)
+    input_program.is_optimized = True
     for cfg in cfgs:
         cfg.memory_optimize(skip_opt_set=skip_opt_set, level=level)
 
@@ -559,5 +560,6 @@ def release_memory(input_program, skip_opt_set=None):
         None
     """
     cfgs = _get_cfgs(input_program)
+    input_program.is_optimized = True
     for cfg in cfgs:
         cfg.release_memory(skip_opt_set=skip_opt_set)