diff --git a/README.md b/README.md
index 375825b..fb816c6 100644
--- a/README.md
+++ b/README.md
@@ -75,4 +75,12 @@ This is also a **SM3** implementation whose performance is similar like golang n
     PASS
     ok  	github.com/emmansun/gmsm/sm3	3.043s    
 
-   
\ No newline at end of file
+**SM4 Benchmark**
+
+    Pure golang version
+    goos: windows
+    goarch: amd64
+    pkg: github.com/emmansun/gmsm/sm4
+    BenchmarkEncrypt-6   	 2671431	       441 ns/op	  36.28 MB/s	       0 B/op	       0 allocs/op
+    BenchmarkDecrypt-6   	 2709132	       440 ns/op	  36.40 MB/s	       0 B/op	       0 allocs/op
+    BenchmarkExpand-6    	 2543746	       471 ns/op	      16 B/op	       1 allocs/op
diff --git a/sm4/cipher_test.go b/sm4/cipher_test.go
index 6767607..e93da01 100644
--- a/sm4/cipher_test.go
+++ b/sm4/cipher_test.go
@@ -10,6 +10,21 @@ import (
 	"testing"
 )
 
+type CryptTest struct {
+	key []byte
+	in  []byte
+	out []byte
+}
+
+var encryptTests = []CryptTest{
+	{
+		// Appendix 1.
+		[]byte{0x01, 0x23, 0x45, 0x67, 0x89, 0xab, 0xcd, 0xef, 0xfe, 0xdc, 0xba, 0x98, 0x76, 0x54, 0x32, 0x10},
+		[]byte{0x01, 0x23, 0x45, 0x67, 0x89, 0xab, 0xcd, 0xef, 0xfe, 0xdc, 0xba, 0x98, 0x76, 0x54, 0x32, 0x10},
+		[]byte{0x68, 0x1e, 0xdf, 0x34, 0xd2, 0x06, 0x96, 0x5e, 0x86, 0xb3, 0xe9, 0x4f, 0x53, 0x6e, 0x42, 0x46},
+	},
+}
+
 func Test_sample1(t *testing.T) {
 	src := []byte{0x01, 0x23, 0x45, 0x67, 0x89, 0xab, 0xcd, 0xef, 0xfe, 0xdc, 0xba, 0x98, 0x76, 0x54, 0x32, 0x10}
 	expected := []byte{0x68, 0x1e, 0xdf, 0x34, 0xd2, 0x06, 0x96, 0x5e, 0x86, 0xb3, 0xe9, 0x4f, 0x53, 0x6e, 0x42, 0x46}
@@ -46,6 +61,43 @@ func Test_sample2(t *testing.T) {
 	}
 }
 
+func BenchmarkEncrypt(b *testing.B) {
+	tt := encryptTests[0]
+	c, err := NewCipher(tt.key)
+	if err != nil {
+		b.Fatal("NewCipher:", err)
+	}
+	out := make([]byte, len(tt.in))
+	b.SetBytes(int64(len(out)))
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		c.Encrypt(out, tt.in)
+	}
+}
+
+func BenchmarkDecrypt(b *testing.B) {
+	tt := encryptTests[0]
+	c, err := NewCipher(tt.key)
+	if err != nil {
+		b.Fatal("NewCipher:", err)
+	}
+	out := make([]byte, len(tt.out))
+	b.SetBytes(int64(len(out)))
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		c.Decrypt(out, tt.out)
+	}
+}
+
+func BenchmarkExpand(b *testing.B) {
+	tt := encryptTests[0]
+	c := &sm4Cipher{make([]uint32, rounds), make([]uint32, rounds)}
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		expandKeyGo(tt.key, c.enc, c.dec)
+	}
+}
+
 func paddingPKCS7(buf []byte, blockSize int) []byte {
 	bufLen := len(buf)
 	padLen := blockSize - bufLen%blockSize