diff --git a/ChangeLog b/ChangeLog
index 6b8b9a4..47f5835 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -166,6 +166,8 @@ it fails when running the interpreter with a 16M stack (and if changing the
 stack size via pcre2test is possible). This avoids having to manually set a
 large stack size when testing with clang.
 
+42. Fix register overwite in JIT when SSE2 acceleration is enabled.
+
 
 Version 10.21 12-January-2016
 -----------------------------
diff --git a/src/pcre2_jit_compile.c b/src/pcre2_jit_compile.c
index c98f977..8dea90a 100644
--- a/src/pcre2_jit_compile.c
+++ b/src/pcre2_jit_compile.c
@@ -4027,12 +4027,12 @@ sljit_emit_op_custom(compiler, instruction, 4);
 
 if (load_twice)
   {
-  OP1(SLJIT_MOV, TMP3, 0, TMP2, 0);
+  OP1(SLJIT_MOV, RETURN_ADDR, 0, TMP2, 0);
   instruction[3] = 0xc0 | (tmp2_ind << 3) | 1;
   sljit_emit_op_custom(compiler, instruction, 4);
 
   OP2(SLJIT_OR, TMP1, 0, TMP1, 0, TMP2, 0);
-  OP1(SLJIT_MOV, TMP2, 0, TMP3, 0);
+  OP1(SLJIT_MOV, TMP2, 0, RETURN_ADDR, 0);
   }
 
 OP2(SLJIT_ASHR, TMP1, 0, TMP1, 0, TMP2, 0);
diff --git a/src/pcre2_jit_test.c b/src/pcre2_jit_test.c
index dfe0032..705ba18 100644
--- a/src/pcre2_jit_test.c
+++ b/src/pcre2_jit_test.c
@@ -692,6 +692,7 @@ static struct regression_test_case regression_test_cases[] = {
 	{ PCRE2_FIRSTLINE | PCRE2_DOTALL, PCRE2_NEWLINE_LF, 0, 0 | F_NOMATCH, "ab.", "ab" },
 	{ MU | PCRE2_FIRSTLINE, A, 0, 1 | F_NOMATCH, "^[a-d0-9]", "\nxx\nd" },
 	{ PCRE2_FIRSTLINE | PCRE2_DOTALL, PCRE2_NEWLINE_ANY, 0, 0, "....a", "012\n0a" },
+	{ MU | PCRE2_FIRSTLINE, A, 0, 0, "[aC]", "a" },
 
 	/* Recurse. */
 	{ MU, A, 0, 0, "(a)(?1)", "aa" },