diff --git a/src/loaders/webp/utils/bit_reader.cpp b/src/loaders/webp/utils/bit_reader.cpp
index a268c646..84c5c46b 100644
--- a/src/loaders/webp/utils/bit_reader.cpp
+++ b/src/loaders/webp/utils/bit_reader.cpp
@@ -16,6 +16,7 @@
 #endif
 
 #include "./bit_reader_inl.h"
+#include "./utils.h"
 
 //------------------------------------------------------------------------------
 // VP8BitReader
@@ -109,11 +110,10 @@ int32_t VP8GetSignedValue(VP8BitReader* const br, int bits) {
 
 #define VP8L_LOG8_WBITS 4  // Number of bytes needed to store VP8L_WBITS bits.
 
-#if !defined(WEBP_FORCE_ALIGNED) && \
-    (defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || \
+#if (defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || \
      defined(__i386__) || defined(_M_IX86) || \
      defined(__x86_64__) || defined(_M_X64))
-#define VP8L_USE_UNALIGNED_LOAD
+#define VP8L_USE_FAST_LOAD
 #endif
 
 static const uint32_t kBitMask[VP8L_MAX_NUM_BIT_READ + 1] = {
@@ -183,13 +183,13 @@ void VP8LDoFillBitWindow(VP8LBitReader* const br) {
   assert(br->bit_pos_ >= VP8L_WBITS);
   // TODO(jzern): given the fixed read size it may be possible to force
   //              alignment in this block.
-#if defined(VP8L_USE_UNALIGNED_LOAD)
+#if defined(VP8L_USE_FAST_LOAD)
   if (br->pos_ + sizeof(br->val_) < br->len_) {
     br->val_ >>= VP8L_WBITS;
     br->bit_pos_ -= VP8L_WBITS;
     // The expression below needs a little-endian arch to work correctly.
     // This gives a large speedup for decoding speed.
-    br->val_ |= (vp8l_val_t)*(const uint32_t*)(br->buf_ + br->pos_) <<
+    br->val_ |= (vp8l_val_t)HToLE32(WebPMemToUint32(br->buf_ + br->pos_)) <<
                 (VP8L_LBITS - VP8L_WBITS);
     br->pos_ += VP8L_LOG8_WBITS;
     return;
diff --git a/src/loaders/webp/utils/bit_reader_inl.h b/src/loaders/webp/utils/bit_reader_inl.h
index 1f79c941..aa317119 100644
--- a/src/loaders/webp/utils/bit_reader_inl.h
+++ b/src/loaders/webp/utils/bit_reader_inl.h
@@ -20,9 +20,7 @@
 #include "../webp/config.h"
 #endif
 
-#ifdef WEBP_FORCE_ALIGNED
 #include <string.h>  // memcpy
-#endif
 
 #include "../dsp/dsp.h"
 #include "./bit_reader.h"
@@ -61,10 +59,7 @@ static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) {
   if (br->buf_ + sizeof(lbit_t) <= br->buf_end_) {
     // convert memory type to register type (with some zero'ing!)
     bit_t bits;
-#if defined(WEBP_FORCE_ALIGNED)
-    lbit_t in_bits;
-    memcpy(&in_bits, br->buf_, sizeof(in_bits));
-#elif defined(WEBP_USE_MIPS32)
+#if defined(WEBP_USE_MIPS32)
     // This is needed because of un-aligned read.
     lbit_t in_bits;
     lbit_t* p_buf_ = (lbit_t*)br->buf_;
@@ -79,7 +74,8 @@ static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) {
       : "memory", "at"
     );
 #else
-    const lbit_t in_bits = *(const lbit_t*)br->buf_;
+    lbit_t in_bits;
+    memcpy(&in_bits, br->buf_, sizeof(in_bits));
 #endif
     br->buf_ += BITS >> 3;
 #if !defined(WORDS_BIGENDIAN)
diff --git a/src/loaders/webp/utils/utils.h b/src/loaders/webp/utils/utils.h
index 3dba9427..29331d23 100644
--- a/src/loaders/webp/utils/utils.h
+++ b/src/loaders/webp/utils/utils.h
@@ -59,6 +59,18 @@ static WEBP_INLINE int BitsLog2Floor(uint32_t n) {
 }
 #endif
 
+//------------------------------------------------------------------------------
+//------------------------------------------------------------------------------
+// Alignment
+
+#include <string.h>
+// memcpy() is the safe way of moving potentially unaligned 32b memory.
+static WEBP_INLINE uint32_t WebPMemToUint32(const uint8_t* const ptr) {
+  uint32_t A;
+  memcpy(&A, ptr, sizeof(A));
+  return A;
+}
+
 //------------------------------------------------------------------------------
 
 #ifdef __cplusplus