diff --git a/Test/objc_msgSend.m b/Test/objc_msgSend.m
index 15f6be8..6270096 100644
--- a/Test/objc_msgSend.m
+++ b/Test/objc_msgSend.m
@@ -3,6 +3,7 @@
 #include <objc/runtime.h>
 #include <assert.h>
 #include <string.h>
+#include <class.h>
 
 id objc_msgSend(id, SEL, ...);
 
@@ -38,6 +39,7 @@ int main(void)
 	TestCls = objc_getClass("Test");
 	objc_msgSend(TestCls, @selector(nothing));
 	objc_msgSend(TestCls, @selector(missing));
+	assert(0 == objc_msgSend(0, @selector(nothing)));
 	id a = objc_msgSend(objc_getClass("Test"), @selector(foo));
 	assert((id)0x42 == a);
 	a = objc_msgSend(TestCls, @selector(foo));
diff --git a/class.h b/class.h
index 9ff3e43..3e2963e 100644
--- a/class.h
+++ b/class.h
@@ -274,7 +274,7 @@ static inline Class classForObject(id obj)
 	{
 		if (sizeof(Class) == 4)
 		{
-			return SmallObjectClasses[1];
+			return SmallObjectClasses[0];
 		}
 		else
 		{
diff --git a/class_table.c b/class_table.c
index 4694444..8feb1f4 100644
--- a/class_table.c
+++ b/class_table.c
@@ -431,6 +431,15 @@ BOOL objc_registerSmallObjectClass_np(Class class, uintptr_t mask)
 	{
 		return NO;
 	}
+	if (sizeof(void*) == 4)
+	{
+		if (Nil == SmallObjectClasses[0])
+		{
+			SmallObjectClasses[0] = class;
+			return YES;
+		}
+		return NO;
+	}
 	if (Nil != SmallObjectClasses[mask])
 	{
 		return NO;
diff --git a/objc_msgSend.S b/objc_msgSend.S
index e171d46..10e52f3 100644
--- a/objc_msgSend.S
+++ b/objc_msgSend.S
@@ -2,7 +2,8 @@
 #include "objc_msgSend.x86-64.S"
 #elif __i386
 #include "objc_msgSend.x86-32.S"
+#elif __arm__
+#include "objc_msgSend.arm.S"
 #else
-//#elif __arm__
 #warning objc_msgSend() not implemented for your architecture
 #endif
diff --git a/objc_msgSend.arm.S b/objc_msgSend.arm.S
new file mode 100644
index 0000000..735c96b
--- /dev/null
+++ b/objc_msgSend.arm.S
@@ -0,0 +1,98 @@
+#define DTABLE_OFFSET  32
+#define SMALLOBJ_MASK  1
+#define SHIFT_OFFSET   4
+#define DATA_OFFSET    12
+#define SLOT_OFFSET    16
+.syntax unified
+
+// Macro for testing: logs a register value to standard error
+.macro LOG reg
+	push {r0-r3, ip,lr}
+	mov r0, \reg
+	bl  logInt(PLT)
+	pop {r0-r3, ip,lr}
+.endm
+
+.macro MSGSEND receiver, sel
+	teq    \receiver, 0
+	beq    4f                              // Skip everything if the receiver is nil
+	push   {r4-r6}                         // We're going to use these three as
+                                           // scratch registers, so save them now
+	tst    \receiver, SMALLOBJ_MASK        // Sets Z if this is not a small int
+
+
+	ldrne  r4, LSmallIntClass              // Small Int class -> r4 if this is a small int
+	ldrne  r4, [r4]
+
+	ldreq  r4, [\receiver]                 // Load class to r4 if not a small int
+
+	ldr    r4, [r4, #DTABLE_OFFSET]        // Dtable -> r4
+
+	ldr    r5, LUninstalledDtable          // &uninstalled_dtable -> r5
+	ldr    r5, [r5]
+
+	teq    r4, r5                          // If dtable == &uninstalled_dtable
+	beq    5f                              // Do a slow lookup
+
+	ldr    r5, [\sel]                      // selector->index -> r5
+
+	ldr    r6, [r4, #SHIFT_OFFSET]        // dtable->shift -> r6
+	ldr    r4, [r4, #DATA_OFFSET]         // dtable->data -> r4
+	
+	teq    r6, #8                         // If this is a small dtable, jump to the small dtable handlers
+	beq    1f
+	teq    r6, #0
+	beq    2f
+
+	and    r6, r5, #0xff0000
+	ldr    r4, [r4, r6, asr#14]
+1:                                        // dtable16
+	and    r6, r5, #0xff00
+	ldr    r4, [r4, r6, asr#6]
+2:                                        // dtable8
+	and    r6, r5, #0xff
+	ldr    ip, [r4, r6, asl#2]
+
+	teq    ip, #0                         // If the slot is nil
+	beq    5f                             // Go to the slow path and do the forwarding stuff
+
+	ldr   ip, [ip, #SLOT_OFFSET]          // Load the method from the slot
+
+3: 
+	pop    {r4-r6}                        // Restore the saved callee-save registers
+	mov    pc, ip
+
+4:                                        // Nil receiver
+	mov    r0, 0
+	mov    r1, 0
+	mov    pc, lr
+5:                                        // Slow lookup
+	push   {r0-r3, lr}                    // Save anything that will be clobbered by the call
+
+	push   {\receiver}                    // &self, _cmd in arguments
+	mov    r0, sp
+	mov    r1, \sel
+	bl     slowMsgLookup(PLT)
+	mov    ip, r0                         // IMP -> ip
+
+	pop    {r5}                           // restore (modified) self to r5
+	pop    {r0-r3, lr}                    // Load clobbered registers
+	mov    \receiver, r5
+	b      3b
+.endm
+
+.globl objc_msgSend
+	.type	objc_msgSend, %function
+objc_msgSend:
+	MSGSEND r0, r1
+.globl objc_msgSend_sret
+	.type	objc_msgSend_sret, %function
+objc_msgSend_sret:
+	MSGSEND r1, r2
+
+LSmallIntClass:
+	.long   SmallObjectClasses
+	.align  2
+LUninstalledDtable:
+	.long   uninstalled_dtable
+	.align  2
diff --git a/objc_msgSend.x86-32.S b/objc_msgSend.x86-32.S
index 8eb6d97..c939f16 100644
--- a/objc_msgSend.x86-32.S
+++ b/objc_msgSend.x86-32.S
@@ -75,7 +75,6 @@
 	call  __i686.get_pc_thunk.bx
 	addl  $_GLOBAL_OFFSET_TABLE_, %ebx
 	mov   SmallObjectClasses@GOT(%ebx), %eax
-	addl  $4, %eax
 	mov   (%eax), %eax
 	popl  %ebx
 	jmp   1b 
diff --git a/sendmsg2.c b/sendmsg2.c
index 1c0d49b..67d7ada 100644
--- a/sendmsg2.c
+++ b/sendmsg2.c
@@ -110,9 +110,9 @@ IMP slowMsgLookup(id *receiver, SEL cmd)
 	return objc_msg_lookup_sender(receiver, cmd, nil)->method;
 }
 
-PRIVATE void logInt(long long a)
+PRIVATE void logInt(void *a)
 {
-	fprintf(stderr, "Value: %llx\n", a);
+	fprintf(stderr, "Value: %p\n", a);
 }
 
 Slot_t (*objc_plane_lookup)(id *receiver, SEL op, id sender) =