diff --git a/.travis.yml b/.travis.yml
index 8dae417..db82257 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -1,2 +1,8 @@
 language: cpp
-script: mkdir build && cd build && cmake -G "Unix Makefiles" .. && make
+script:
+ - mkdir build
+ - cd build
+ - cmake -G "Unix Makefiles" ..
+ - make
+ - export PATH=$(pwd):$PATH
+ - ../tests/test.sh
diff --git a/core/src/zxing/common/StringUtils.cpp b/core/src/zxing/common/StringUtils.cpp
index 30f9325..e5cea76 100644
--- a/core/src/zxing/common/StringUtils.cpp
+++ b/core/src/zxing/common/StringUtils.cpp
@@ -25,7 +25,9 @@ using namespace zxing::common;
 
 // N.B.: these are the iconv strings for at least some versions of iconv
 
-char const* const StringUtils::PLATFORM_DEFAULT_ENCODING = "UTF-8";
+char const* const StringUtils::PLATFORM_DEFAULT_ENCODING = "ISO8859-1";
+  // This default is an 8-bit fixed-width encoding and it's identical with
+  // unicode in its range, which reduces accidental mangling during decode.
 char const* const StringUtils::ASCII = "ASCII";
 char const* const StringUtils::SHIFT_JIS = "SHIFT_JIS";
 char const* const StringUtils::GB2312 = "GBK";
@@ -76,6 +78,11 @@ StringUtils::guessEncoding(char* bytes, int length,
 
     int value = bytes[i] & 0xFF;
 
+    // embedded nuls are a sure sign of binary data
+    if(value == '\x00') {
+      canBeISO88591 = canBeShiftJIS = canBeUTF8 = false;
+    }
+
     // UTF-8 stuff
     if (canBeUTF8) {
       if (utf8BytesLeft > 0) {
diff --git a/core/src/zxing/qrcode/decoder/DecodedBitStreamParser.cpp b/core/src/zxing/qrcode/decoder/DecodedBitStreamParser.cpp
index 1d36bba..40f5c32 100644
--- a/core/src/zxing/qrcode/decoder/DecodedBitStreamParser.cpp
+++ b/core/src/zxing/qrcode/decoder/DecodedBitStreamParser.cpp
@@ -67,6 +67,12 @@ void DecodedBitStreamParser::append(std::string &result,
     return;
   }
 
+  if(src == NULL) {
+    // don't try to recode un-encoded data.
+    result.append((const char *)bufIn, nIn);
+    return;
+  }
+
   iconv_t cd = iconv_open(StringUtils::UTF8, src);
   if (cd == (iconv_t)-1) {
     result.append((const char *)bufIn, nIn);
@@ -193,19 +199,24 @@ void DecodedBitStreamParser::decodeByteSegment(Ref<BitSource> bits_,
   for (int i = 0; i < count; i++) {
     readBytes[i] = (char) bits.readBits(8);
   }
-  string encoding;
-  if (currentCharacterSetECI == 0) {
-    // The spec isn't clear on this mode; see
-    // section 6.4.5: t does not say which encoding to assuming
-    // upon decoding. I have seen ISO-8859-1 used as well as
-    // Shift_JIS -- without anything like an ECI designator to
-    // give a hint.
-    encoding = StringUtils::guessEncoding(readBytes, count, hints);
+  const char* encoding = NULL;
+  if (currentCharacterSetECI == NULL) {
+    // The spec says
+    // 8.3.1: The default interpretation for QR Code is ECI 000020 representing the JIS8 and Shift JIS character sets.
+    // 8.4.4: In [8-bit Byte Mode], one 8 bit codeword directly represents the JIS8 character [...].
+    //        In ECIs other than the default ECI, it represents an 8-bit byte value directly.
+    // If I'm reading that right, *if* the character set is unspecified *or* explicitly set to Shift-JIS,
+    // *then* use JIS8 but *otherwise* don't try to decode the value.
+    //
+    // That's a stupid spec.
+    // Instead we follow qrencode:
+    //   unspecified ECI <=> unmolested binary
+    encoding = NULL;
   } else {
     encoding = currentCharacterSetECI->name();
   }
   try {
-    append(result, readBytes, nBytes, encoding.c_str());
+    append(result, readBytes, nBytes, encoding);
   } catch (ReaderException const& ignored) {
     (void)ignored;
     throw FormatException();
diff --git a/tests/amen-01.bin b/tests/amen-01.bin
new file mode 100644
index 0000000..74d21bf
Binary files /dev/null and b/tests/amen-01.bin differ
diff --git a/tests/amen-01.png b/tests/amen-01.png
new file mode 100644
index 0000000..599f494
Binary files /dev/null and b/tests/amen-01.png differ
diff --git a/tests/test.sh b/tests/test.sh
new file mode 100755
index 0000000..3c7e825
--- /dev/null
+++ b/tests/test.sh
@@ -0,0 +1,10 @@
+#!/bin/bash
+
+
+cd "$(dirname "$0")";
+
+find . -name "*.png" | while read file; do
+  #zxing has a --test-mode but it's not working for binary files
+  # perl is here because in textmode, zxing appends a newline, but a lot of test files don't have that
+  zxing "$file" | perl -pe 'chomp if eof' | diff -u "$(ls "${file%.*}".{txt,bin} 2>/dev/null)" -
+done