diff --git a/benchmark/README.md b/benchmark/README.md
new file mode 100644
index 000000000..1c9f1918c
--- /dev/null
+++ b/benchmark/README.md
@@ -0,0 +1,32 @@
+benchncnn can be used to test neural network inference performance
+
+Only the network definition files (ncnn param) are required.
+
+The large model binary files (ncnn bin) are not loaded but generated randomly for speed test.
+
+More model networks may be added later.
+
+---
+
+Usage
+```
+# copy all param files to the current directory
+./benchncnn [loop count] [num threads] [powersave]
+```
+
+Typical output (executed in android adb shell)
+```
+HM2014812:/data/local/tmp # ./benchncnn 4 4 0
+loop_count = 4
+num_threads = 4
+powersave = 0
+      squeezenet  min =   99.34  max =  104.24  avg =  102.83
+       mobilenet  min =  171.06  max =  177.08  avg =  173.45
+      shufflenet  min =   67.80  max =   73.63  avg =   69.35
+       googlenet  min =  337.30  max =  341.33  avg =  339.00
+        resnet18  min =  466.41  max =  476.33  avg =  472.07
+         alexnet  min =  397.47  max =  421.54  avg =  404.22
+           vgg16  min = 2338.67  max = 2984.17  avg = 2557.16
+  squeezenet-ssd  min =  201.43  max =  304.06  avg =  236.67
+   mobilenet-ssd  min =  198.93  max =  241.33  avg =  214.54
+```