diff --git a/doc/SAVE.md b/doc/SAVE.md
index c1e6b19a45c75a64207802984f52c734d44f8fc8..3f7f97e12e1e309ff0933e150ea7bcd23298b60e 100644
--- a/doc/SAVE.md
+++ b/doc/SAVE.md
@@ -1,8 +1,9 @@
-## How to save a servable model of Paddle Serving?
+# How to save a servable model of Paddle Serving?
 
 ([简体中文](./SAVE_CN.md)|English)
 
-- Currently, paddle serving provides a save_model interface for users to access, the interface is similar with `save_inference_model` of Paddle.
+## Save from training or prediction script 
+Currently, paddle serving provides a save_model interface for users to access, the interface is similar with `save_inference_model` of Paddle.
 ``` python
 import paddle_serving_client.io as serving_io
 serving_io.save_model("imdb_model", "imdb_client_conf",
@@ -29,3 +30,15 @@ for line in sys.stdin:
     fetch_map = client.predict(feed=feed, fetch=fetch)
     print("{} {}".format(fetch_map["prediction"][1], label[0]))
  ```
+
+## Export from saved model files
+If you have saved model files using Paddle's `save_inference_model` API, you can use Paddle Serving's` inference_model_to_serving` API to convert it into a model file that can be used for Paddle Serving.
+```
+import paddle_serving_client.io as serving_io
+serving_io.inference_model_to_serving(dirname, model_filename=None, params_filename=None, serving_server="serving_server", serving_client="serving_client")
+```
+dirname (str) - Path of saved model files. Program file and parameter files are saved in this directory.
+model_filename (str, optional) - The name of file to load the inference program. If it is None, the default filename __model__ will be used. Default: None.
+paras_filename (str, optional) - The name of file to load all parameters. It is only used for the case that all parameters were saved in a single binary file. If parameters were saved in separate files, set it as None. Default: None.
+serving_server (str, optional) - The path of model files and configuration files for server. Default: "serving_server".
+serving_client (str, optional) - The path of configuration files for client. Default: "serving_client".
diff --git a/doc/SAVE_CN.md b/doc/SAVE_CN.md
index 43b62c2ac623b386505356194ac136ea305fe683..fc75cd8d015a6d6f42a08f29e4035db20f450d91 100644
--- a/doc/SAVE_CN.md
+++ b/doc/SAVE_CN.md
@@ -1,8 +1,9 @@
-## 怎样保存用于Paddle Serving的模型？
+# 怎样保存用于Paddle Serving的模型？
 
 (简体中文|[English](./SAVE.md))
 
-- 目前，Paddle Serving提供了一个save_model接口供用户访问，该接口与Paddle的`save_inference_model`类似。
+## 从训练或预测脚本中保存
+目前，Paddle Serving提供了一个save_model接口供用户访问，该接口与Paddle的`save_inference_model`类似。
 
 ``` python
 import paddle_serving_client.io as serving_io
@@ -29,3 +30,15 @@ for line in sys.stdin:
     fetch_map = client.predict(feed=feed, fetch=fetch)
     print("{} {}".format(fetch_map["prediction"][1], label[0]))
  ```
+
+## 从已保存的模型文件中导出
+如果已使用Paddle 的`save_inference_model`接口保存出预测要使用的模型，则可以通过Paddle Serving的`inference_model_to_serving`接口转换成可用于Paddle Serving的模型文件。
+```
+import paddle_serving_client.io as serving_io
+serving_io.inference_model_to_serving(dirname, model_filename=None, params_filename=None, serving_server="serving_server", serving_client="serving_client")
+```
+dirname (str) – 需要转换的模型文件存储路径，Program结构文件和参数文件均保存在此目录。
+model_filename (str，可选) – 存储需要转换的模型Inference Program结构的文件名称。如果设置为None，则使用 __model__ 作为默认的文件名。默认值为None。
+params_filename (str，可选) – 存储需要转换的模型所有参数的文件名称。当且仅当所有模型参数被保存在一个单独的二进制文件中，它才需要被指定。如果模型参数是存储在各自分离的文件中，设置它的值为None。默认值为None。
+serving_server (str, 可选) - 转换后的模型文件和配置文件的存储路径。默认值为"serving_server"。
+serving_client (str, 可选) - 转换后的客户端配置文件存储路径。默认值为"serving_client"。
diff --git a/python/paddle_serving_client/io/__init__.py b/python/paddle_serving_client/io/__init__.py
index 74a6ca871b5c1e32b3c1ecbc6656c95d7c78a399..4f174866e5521577ba35f39216f7dd0793879a6c 100644
--- a/python/paddle_serving_client/io/__init__.py
+++ b/python/paddle_serving_client/io/__init__.py
@@ -103,17 +103,21 @@ def save_model(server_model_folder,
         fout.write(config.SerializeToString())
 
 
-def inference_model_to_serving(infer_model, serving_client, serving_server):
+def inference_model_to_serving(dirname,
+                               model_filename=None,
+                               params_filename=None,
+                               serving_server="serving_server",
+                               serving_client="serving_client"):
     place = fluid.CPUPlace()
     exe = fluid.Executor(place)
     inference_program, feed_target_names, fetch_targets = \
-            fluid.io.load_inference_model(dirname=infer_model, executor=exe)
+            fluid.io.load_inference_model(dirname=dirname, executor=exe, model_filename=model_filename, params_filename=params_filename)
     feed_dict = {
         x: inference_program.global_block().var(x)
         for x in feed_target_names
     }
     fetch_dict = {x.name: x for x in fetch_targets}
-    save_model(serving_client, serving_server, feed_dict, fetch_dict,
+    save_model(serving_server, serving_client, feed_dict, fetch_dict,
                inference_program)
     feed_names = feed_dict.keys()
     fetch_names = fetch_dict.keys()