diff --git a/demo/mnist/api_train.py b/demo/mnist/api_train.py
new file mode 100644
index 0000000000000000000000000000000000000000..6abb5d4e562ee9d24a1f914757f3b8e4a3e5cb12
--- /dev/null
+++ b/demo/mnist/api_train.py
@@ -0,0 +1,12 @@
+import py_paddle.swig_paddle as api
+from paddle.trainer.config_parser import parse_config
+
+
+def main():
+    api.initPaddle("-use_gpu=false", "-trainer_count=4")  # use 4 cpu cores
+    config = parse_config('simple_mnist_network.py', '')
+    m = api.GradientMachine.createFromConfigProto(config.model_config)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/demo/mnist/simple_mnist_network.py b/demo/mnist/simple_mnist_network.py
new file mode 100644
index 0000000000000000000000000000000000000000..41f4e51657d35bf72401f4076c53b7b3bf7d5b52
--- /dev/null
+++ b/demo/mnist/simple_mnist_network.py
@@ -0,0 +1,16 @@
+from paddle.trainer_config_helpers import *
+
+settings(learning_rate=1e-4, learning_method=AdamOptimizer(), batch_size=1000)
+
+imgs = data_layer(name='pixel', size=784)
+
+hidden1 = fc_layer(input=imgs, size=200)
+hidden2 = fc_layer(input=hidden1, size=200)
+
+inference = fc_layer(input=hidden2, size=10, act=SoftmaxActivation())
+
+cost = classification_cost(
+    input=inference, label=data_layer(
+        name='label', size=10))
+
+outputs(cost)