diff --git a/re2/parse.cc b/re2/parse.cc
index 9045992..a027917 100644
--- a/re2/parse.cc
+++ b/re2/parse.cc
@@ -776,7 +776,8 @@
   while (re->op() == kRegexpConcat && re->nsub() > 0)
     re = re->sub()[0];
 
-  *flags = static_cast<Regexp::ParseFlags>(re->parse_flags_ & Regexp::FoldCase);
+  *flags = static_cast<Regexp::ParseFlags>(re->parse_flags_ &
+                                           (Regexp::FoldCase | Regexp::Latin1));
 
   if (re->op() == kRegexpLiteral) {
     *nrune = 1;
diff --git a/re2/testing/re2_test.cc b/re2/testing/re2_test.cc
index 151525f..ddf8dbf 100644
--- a/re2/testing/re2_test.cc
+++ b/re2/testing/re2_test.cc
@@ -1658,4 +1658,23 @@
   ASSERT_EQ(m, "") << " got m='" << m << "', want ''";
 }
 
+TEST(RE2, Issue477) {
+  // Regexp::LeadingString didn't output Latin1 into flags.
+  // In the given pattern, 0xA5 should be factored out, but
+  // shouldn't lose its Latin1-ness in the process. Because
+  // that was happening, the prefix for accel was 0xC2 0xA5
+  // instead of 0xA5. Note that the former doesn't occur in
+  // the given input and so replacements weren't occurring.
+
+  const char bytes[] = {
+      (char)0xa5, (char)0xd1, (char)0xa5, (char)0xd1,
+      (char)0x61, (char)0x63, (char)0xa5, (char)0x64,
+  };
+  std::string s(bytes, ABSL_ARRAYSIZE(bytes));
+  RE2 re("\xa5\xd1|\xa5\x64", RE2::Latin1);
+  int n = RE2::GlobalReplace(&s, re, "");
+  ASSERT_EQ(n, 3);
+  ASSERT_EQ(s, "\x61\x63");
+}
+
 }  // namespace re2