diff --git a/README.md b/README.md
index fb7ac3453..ffed0b258 100644
--- a/README.md
+++ b/README.md
@@ -95,6 +95,10 @@ building the library or including the function definitions:
 #define BOOST_JSON_STACK_BUFFER_SIZE 1024
 #include <boost/json/src.hpp>
 ```
+### Endianness
+
+Boost.JSON uses [Boost.Endian](https://www.boost.org/doc/libs/release/libs/endian/doc/html/endian.html)
+in order to support both little endian and big endian platforms.
 
 ### Supported Compilers
 
diff --git a/doc/qbk/overview.qbk b/doc/qbk/overview.qbk
index 9de3cd724..d4f564b8d 100644
--- a/doc/qbk/overview.qbk
+++ b/doc/qbk/overview.qbk
@@ -120,6 +120,12 @@ building the library or including the function definitions:
 #include <boost/json/src.hpp>
 ```
 
+[heading Endianness]
+
+Boost.JSON uses
+[@https://www.boost.org/doc/libs/release/libs/endian/doc/html/endian.html
+Boost.Endian] in order to support both little endian and big endian platforms.
+
 [heading Supported Compilers]
 
 Boost.JSON has been tested with the following compilers:
diff --git a/include/boost/json/detail/sse2.hpp b/include/boost/json/detail/sse2.hpp
index 06657bc18..3b979c467 100644
--- a/include/boost/json/detail/sse2.hpp
+++ b/include/boost/json/detail/sse2.hpp
@@ -136,7 +136,7 @@ count_valid<false>(
         uint8_t len = first & 0xFF;
         if(BOOST_JSON_UNLIKELY(end - p < len))
             break;
-        if(BOOST_JSON_UNLIKELY(! is_valid_utf8(p, first)))
+        if(BOOST_JSON_UNLIKELY(! is_valid_utf8_no_inline(p, first)))
             break;
         p += len;
     }
@@ -185,7 +185,7 @@ count_valid<false>(
         uint8_t len = first & 0xFF;
         if(BOOST_JSON_UNLIKELY(end - p < len))
             break;
-        if(BOOST_JSON_UNLIKELY(! is_valid_utf8(p, first)))
+        if(BOOST_JSON_UNLIKELY(! is_valid_utf8_no_inline(p, first)))
             break;
         p += len;
     }
diff --git a/include/boost/json/detail/utf8.hpp b/include/boost/json/detail/utf8.hpp
index 74434773e..0a0fea2b2 100644
--- a/include/boost/json/detail/utf8.hpp
+++ b/include/boost/json/detail/utf8.hpp
@@ -25,10 +25,8 @@ template<int N>
 std::uint32_t
 load_little_endian(void const* p)
 {
-    std::uint32_t v = 0;
-    std::memcpy(&v, p, N);
-    endian::little_to_native_inplace(v);
-    return v;
+    auto const up = reinterpret_cast<unsigned char const*>(p);
+    return endian::endian_load<std::uint32_t, N, endian::order::little>(up);
 }
 
 inline
@@ -70,7 +68,7 @@ inline
 bool
 is_valid_utf8(const char* p, uint16_t first)
 {
-    uint32_t v;
+    std::uint32_t v;
     switch(first >> 8)
     {
     default:
@@ -81,38 +79,46 @@ is_valid_utf8(const char* p, uint16_t first)
         v = load_little_endian<2>(p);
         return (v & 0xC000) == 0x8000;
 
-    // 3 bytes, second byte [A0, BF]
-    case 2:
+     // 3 bytes, second byte [A0, BF]
+     case 2:
         v = load_little_endian<3>(p);
         return (v & 0xC0E000) == 0x80A000;
 
-    // 3 bytes, second byte [80, BF]
-    case 3:
+     // 3 bytes, second byte [80, BF]
+     case 3:
         v = load_little_endian<3>(p);
         return (v & 0xC0C000) == 0x808000;
 
-    // 3 bytes, second byte [80, 9F]
-    case 4:
+     // 3 bytes, second byte [80, 9F]
+     case 4:
         v = load_little_endian<3>(p);
         return (v & 0xC0E000) == 0x808000;
 
-    // 4 bytes, second byte [90, BF]
-    case 5:
+     // 4 bytes, second byte [90, BF]
+     case 5:
         v = load_little_endian<4>(p);
         return (v & 0xC0C0FF00) + 0x7F7F7000 <= 0x2F00;
 
-    // 4 bytes, second byte [80, BF]
-    case 6:
+     // 4 bytes, second byte [80, BF]
+     case 6:
         v = load_little_endian<4>(p);
         return (v & 0xC0C0C000) == 0x80808000;
 
-    // 4 bytes, second byte [80, 8F]
-    case 7:
+     // 4 bytes, second byte [80, 8F]
+     case 7:
         v = load_little_endian<4>(p);
         return (v & 0xC0C0F000) == 0x80808000;
     }
 }
 
+BOOST_NOINLINE
+inline
+bool
+is_valid_utf8_no_inline(const char* p, uint16_t first)
+{
+    return is_valid_utf8(p, first);
+}
+
 class utf8_sequence
 {
     char seq_[4];