diff --git a/test/minmax_test.cc b/test/minmax_test.cc
new file mode 100644
index 000000000..dbe4342dc
--- /dev/null
+++ b/test/minmax_test.cc
@@ -0,0 +1,132 @@
+/*
+ *  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+#include "test/acm_random.h"
+#include "test/register_state_check.h"
+
+namespace {
+
+using ::libvpx_test::ACMRandom;
+
+typedef void (*MinMaxFunc)(const uint8_t *a, int a_stride,
+                           const uint8_t *b, int b_stride,
+                           int *min, int *max);
+
+class MinMaxTest : public ::testing::TestWithParam<MinMaxFunc> {
+ public:
+  virtual void SetUp() {
+    mm_func_ = GetParam();
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+  }
+
+ protected:
+  MinMaxFunc mm_func_;
+  ACMRandom rnd_;
+};
+
+void reference_minmax(const uint8_t *a, int a_stride,
+                      const uint8_t *b, int b_stride,
+                      int *min_ret, int *max_ret) {
+  int min = 255;
+  int max = 0;
+  for (int i = 0; i < 8; i++) {
+    for (int j = 0; j < 8; j++) {
+      const int diff = abs(a[i * a_stride + j] - b[i * b_stride + j]);
+      if (min > diff) min = diff;
+      if (max < diff) max = diff;
+    }
+  }
+
+  *min_ret = min;
+  *max_ret = max;
+}
+
+TEST_P(MinMaxTest, MinValue) {
+  for (int i = 0; i < 64; i++) {
+    uint8_t a[64], b[64];
+    memset(a, 0, sizeof(a));
+    memset(b, 255, sizeof(b));
+    b[i] = i;  // Set a minimum difference of i.
+
+    int min, max;
+    ASM_REGISTER_STATE_CHECK(mm_func_(a, 8, b, 8, &min, &max));
+    EXPECT_EQ(255, max);
+    EXPECT_EQ(i, min);
+  }
+}
+
+TEST_P(MinMaxTest, MaxValue) {
+  for (int i = 0; i < 64; i++) {
+    uint8_t a[64], b[64];
+    memset(a, 0, sizeof(a));
+    memset(b, 0, sizeof(b));
+    b[i] = i;  // Set a maximum difference of i.
+
+    int min, max;
+    ASM_REGISTER_STATE_CHECK(mm_func_(a, 8, b, 8, &min, &max));
+    EXPECT_EQ(i, max);
+    EXPECT_EQ(0, min);
+  }
+}
+
+TEST_P(MinMaxTest, CompareReference) {
+  uint8_t a[64], b[64];
+  for (int j = 0; j < 64; j++) {
+    a[j] = rnd_.Rand8();
+    b[j] = rnd_.Rand8();
+  }
+
+  int min_ref, max_ref, min, max;
+  reference_minmax(a, 8, b, 8, &min_ref, &max_ref);
+  ASM_REGISTER_STATE_CHECK(mm_func_(a, 8, b, 8, &min, &max));
+  EXPECT_EQ(max_ref, max);
+  EXPECT_EQ(min_ref, min);
+}
+
+TEST_P(MinMaxTest, CompareReferenceAndVaryStride) {
+  uint8_t a[8 * 64], b[8 * 64];
+  for (int i = 0; i < 8 * 64; i++) {
+    a[i] = rnd_.Rand8();
+    b[i] = rnd_.Rand8();
+  }
+  for (int a_stride = 8; a_stride <= 64; a_stride += 8) {
+    for (int b_stride = 8; b_stride <= 64; b_stride += 8) {
+      int min_ref, max_ref, min, max;
+      reference_minmax(a, a_stride, b, b_stride, &min_ref, &max_ref);
+      ASM_REGISTER_STATE_CHECK(mm_func_(a, a_stride, b, b_stride, &min, &max));
+      EXPECT_EQ(max_ref, max) << "when a_stride = " << a_stride
+                              << " and b_stride = " << b_stride;;
+      EXPECT_EQ(min_ref, min) << "when a_stride = " << a_stride
+                              << " and b_stride = " << b_stride;;
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, MinMaxTest, ::testing::Values(&vpx_minmax_8x8_c));
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, MinMaxTest,
+                        ::testing::Values(&vpx_minmax_8x8_sse2));
+#endif
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, MinMaxTest,
+                        ::testing::Values(&vpx_minmax_8x8_neon));
+#endif
+
+}  // namespace
diff --git a/test/test.mk b/test/test.mk
index d28ab114b..7c22ca501 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -144,6 +144,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += dct32x32_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct4x4_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += hadamard_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += minmax_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_error_block_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_quantize_test.cc
diff --git a/vpx_dsp/arm/avg_neon.c b/vpx_dsp/arm/avg_neon.c
index d054c4185..e52958c54 100644
--- a/vpx_dsp/arm/avg_neon.c
+++ b/vpx_dsp/arm/avg_neon.c
@@ -197,3 +197,60 @@ int vpx_vector_var_neon(int16_t const *ref, int16_t const *src, const int bwl) {
     return s - ((t * t) >> shift_factor);
   }
 }
+
+void vpx_minmax_8x8_neon(const uint8_t *a, int a_stride,
+                         const uint8_t *b, int b_stride,
+                         int *min, int *max) {
+  // Load and concatenate.
+  const uint8x16_t a01 = vcombine_u8(vld1_u8(a),
+                                     vld1_u8(a + a_stride));
+  const uint8x16_t a23 = vcombine_u8(vld1_u8(a + 2 * a_stride),
+                                     vld1_u8(a + 3 * a_stride));
+  const uint8x16_t a45 = vcombine_u8(vld1_u8(a + 4 * a_stride),
+                                     vld1_u8(a + 5 * a_stride));
+  const uint8x16_t a67 = vcombine_u8(vld1_u8(a + 6 * a_stride),
+                                     vld1_u8(a + 7 * a_stride));
+
+  const uint8x16_t b01 = vcombine_u8(vld1_u8(b),
+                                     vld1_u8(b + b_stride));
+  const uint8x16_t b23 = vcombine_u8(vld1_u8(b + 2 * b_stride),
+                                     vld1_u8(b + 3 * b_stride));
+  const uint8x16_t b45 = vcombine_u8(vld1_u8(b + 4 * b_stride),
+                                     vld1_u8(b + 5 * b_stride));
+  const uint8x16_t b67 = vcombine_u8(vld1_u8(b + 6 * b_stride),
+                                     vld1_u8(b + 7 * b_stride));
+
+  // Absolute difference.
+  const uint8x16_t ab01_diff = vabdq_u8(a01, b01);
+  const uint8x16_t ab23_diff = vabdq_u8(a23, b23);
+  const uint8x16_t ab45_diff = vabdq_u8(a45, b45);
+  const uint8x16_t ab67_diff = vabdq_u8(a67, b67);
+
+  // Max values between the Q vectors.
+  const uint8x16_t ab0123_max = vmaxq_u8(ab01_diff, ab23_diff);
+  const uint8x16_t ab4567_max = vmaxq_u8(ab45_diff, ab67_diff);
+  const uint8x16_t ab0123_min = vminq_u8(ab01_diff, ab23_diff);
+  const uint8x16_t ab4567_min = vminq_u8(ab45_diff, ab67_diff);
+
+  const uint8x16_t ab07_max = vmaxq_u8(ab0123_max, ab4567_max);
+  const uint8x16_t ab07_min = vminq_u8(ab0123_min, ab4567_min);
+
+  // Split to D and start doing pairwise.
+  uint8x8_t ab_max = vmax_u8(vget_high_u8(ab07_max), vget_low_u8(ab07_max));
+  uint8x8_t ab_min = vmin_u8(vget_high_u8(ab07_min), vget_low_u8(ab07_min));
+
+  // Enough runs of vpmax/min propogate the max/min values to every position.
+  ab_max = vpmax_u8(ab_max, ab_max);
+  ab_min = vpmin_u8(ab_min, ab_min);
+
+  ab_max = vpmax_u8(ab_max, ab_max);
+  ab_min = vpmin_u8(ab_min, ab_min);
+
+  ab_max = vpmax_u8(ab_max, ab_max);
+  ab_min = vpmin_u8(ab_min, ab_min);
+
+  *min = *max = 0;  // Clear high bits
+  // Store directly to avoid costly neon->gpr transfer.
+  vst1_lane_u8((uint8_t *)max, ab_max, 0);
+  vst1_lane_u8((uint8_t *)min, ab_min, 0);
+}
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 2b131929e..9ea80a098 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -1014,7 +1014,7 @@ if ((vpx_config("CONFIG_VP9_ENCODER") eq "yes") || (vpx_config("CONFIG_VP10_ENCO
   specialize qw/vpx_avg_4x4 sse2 neon msa/;
 
   add_proto qw/void vpx_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max";
-  specialize qw/vpx_minmax_8x8 sse2/;
+  specialize qw/vpx_minmax_8x8 sse2 neon/;
 
   add_proto qw/void vpx_hadamard_8x8/, "const int16_t *src_diff, int src_stride, int16_t *coeff";
   specialize qw/vpx_hadamard_8x8 sse2/, "$ssse3_x86_64_x86inc";