diff --git a/if1632/relay.c b/if1632/relay.c
index 9e4e304a0d3..38df3abd153 100644
--- a/if1632/relay.c
+++ b/if1632/relay.c
@@ -67,10 +67,10 @@ WORD CALLBACK CallTo16Word( FARPROC16 target, INT nArgs )
 LONG CALLBACK CallTo16Long( FARPROC16 target, INT nArgs )
 { assert( FALSE ); }
 
-LONG CALLBACK CallTo16RegisterShort( const CONTEXT86 *context, INT nArgs )
+void CALLBACK CallTo16RegisterShort( CONTEXT86 *context, INT nArgs )
 { assert( FALSE ); }
 
-LONG CALLBACK CallTo16RegisterLong ( const CONTEXT86 *context, INT nArgs )
+void CALLBACK CallTo16RegisterLong ( CONTEXT86 *context, INT nArgs )
 { assert( FALSE ); }
 
 WORD CallFrom16Word( void )       
@@ -327,12 +327,27 @@ void RELAY_DebugCallTo16( LPVOID target, int nb_args, BOOL reg_func )
 /***********************************************************************
  *           RELAY_DebugCallTo16Ret
  */
-void RELAY_DebugCallTo16Ret( int ret_val )
+void RELAY_DebugCallTo16Ret( BOOL reg_func, int ret_val )
 {
     if (!TRACE_ON(relay)) return;
 
-    DPRINTF("CallTo16() ss:sp=%04x:%04x retval=0x%08x\n", 
-            SELECTOROF(NtCurrentTeb()->cur_stack),
-            OFFSETOF(NtCurrentTeb()->cur_stack), ret_val);
+    if (!reg_func)
+    {
+        DPRINTF("CallTo16() ss:sp=%04x:%04x retval=0x%08x\n", 
+                SELECTOROF(NtCurrentTeb()->cur_stack),
+                OFFSETOF(NtCurrentTeb()->cur_stack), ret_val);
+    }
+    else
+    {
+        CONTEXT86 *context = (CONTEXT86 *)ret_val;
+
+        DPRINTF("CallTo16() ss:sp=%04x:%04x\n", 
+                SELECTOROF(NtCurrentTeb()->cur_stack),
+                OFFSETOF(NtCurrentTeb()->cur_stack));
+        DPRINTF("     AX=%04x BX=%04x CX=%04x DX=%04x BP=%04x SP=%04x\n",
+                AX_reg(context), BX_reg(context), CX_reg(context),
+                DX_reg(context), BP_reg(context), LOWORD(ESP_reg(context)));
+    }
+
     SYSLEVEL_CheckNotLevel( 2 );
 }
diff --git a/include/builtin16.h b/include/builtin16.h
index e069ad8873b..1ffe173b4c9 100644
--- a/include/builtin16.h
+++ b/include/builtin16.h
@@ -20,8 +20,8 @@ extern void CallFrom16Thunk();
 
 extern WORD CALLBACK CallTo16Word( FARPROC16 target, INT nArgs );
 extern LONG CALLBACK CallTo16Long( FARPROC16 target, INT nArgs );
-extern LONG CALLBACK CallTo16RegisterShort( const struct _CONTEXT86 *context, INT nArgs );
-extern LONG CALLBACK CallTo16RegisterLong ( const struct _CONTEXT86 *context, INT nArgs );
+extern void CALLBACK CallTo16RegisterShort( struct _CONTEXT86 *context, INT nArgs );
+extern void CALLBACK CallTo16RegisterLong ( struct _CONTEXT86 *context, INT nArgs );
 
 #include "pshpack1.h"
 
diff --git a/loader/ne/module.c b/loader/ne/module.c
index 590a5f97918..5bbceb5e289 100644
--- a/loader/ne/module.c
+++ b/loader/ne/module.c
@@ -1193,7 +1193,8 @@ static void NE_InitProcess(void)
               SELECTOROF(pTask->teb->cur_stack),
               OFFSETOF(pTask->teb->cur_stack) );
 
-        ExitThread( CallTo16RegisterShort( &context, 0 ) );
+        CallTo16RegisterShort( &context, 0 );
+        ExitThread( AX_reg( &context ) );
     }
 
     SYSLEVEL_LeaveWin16Lock();
diff --git a/tools/winebuild/relay.c b/tools/winebuild/relay.c
index fdab80cc28d..018dc45d6f3 100644
--- a/tools/winebuild/relay.c
+++ b/tools/winebuild/relay.c
@@ -504,20 +504,44 @@ static void BuildCallTo16Core( FILE *outfile, int short_ret, int reg_func )
     fprintf( outfile, "\tpushl %%cs\n" );
     fprintf( outfile, "\tcall .LCallTo16%s\n", name );
 
-    /* Convert and push return value */
-    if ( short_ret )
+    if ( !reg_func )
     {
-        fprintf( outfile, "\tmovzwl %%ax, %%eax\n" );
-        fprintf( outfile, "\tpushl %%eax\n" );
-    }
-    else if ( reg_func != 2 )
-    {
-        fprintf( outfile, "\tshll $16,%%edx\n" );
-        fprintf( outfile, "\tmovw %%ax,%%dx\n" );
-        fprintf( outfile, "\tpushl %%edx\n" );
+        /* Convert and push return value */
+        if ( short_ret )
+        {
+            fprintf( outfile, "\tmovzwl %%ax, %%eax\n" );
+            fprintf( outfile, "\tpushl %%eax\n" );
+        }
+        else
+        {
+            fprintf( outfile, "\tshll $16,%%edx\n" );
+            fprintf( outfile, "\tmovw %%ax,%%dx\n" );
+            fprintf( outfile, "\tpushl %%edx\n" );
+        }
     }
     else
-        fprintf( outfile, "\tpushl %%eax\n" );
+    {
+        /* 
+         * Modify CONTEXT86 structure to contain new values
+         *
+         * NOTE:  We restore only EAX, EBX, EDX, EDX, EBP, and ESP.
+         *        The segment registers as well as ESI and EDI should
+         *        not be modified by a well-behaved 16-bit routine in
+         *        any case.  [If necessary, we could restore them as well,
+         *        at the cost of a somewhat less efficient return path.]
+         */
+        
+        fprintf( outfile, "\tmovl %d(%%esp), %%edi\n", STACK32OFFSET(target)-12 );
+        fprintf( outfile, "\tmovl %%eax, %d(%%edi)\n", CONTEXTOFFSET(Eax) );
+        fprintf( outfile, "\tmovl %%ebx, %d(%%edi)\n", CONTEXTOFFSET(Ebx) );
+        fprintf( outfile, "\tmovl %%ecx, %d(%%edi)\n", CONTEXTOFFSET(Ecx) );
+        fprintf( outfile, "\tmovl %%edx, %d(%%edi)\n", CONTEXTOFFSET(Edx) );
+        fprintf( outfile, "\tmovl %%ebp, %d(%%edi)\n", CONTEXTOFFSET(Ebp) );
+        fprintf( outfile, "\tmovl %%esi, %d(%%edi)\n", CONTEXTOFFSET(Esp) );
+                 /* The return glue code saved %esp into %esi */
+
+        fprintf( outfile, "\tpushl %%edi\n" );
+    }
 
     if ( UsePIC )
     {
@@ -531,10 +555,14 @@ static void BuildCallTo16Core( FILE *outfile, int short_ret, int reg_func )
     /* Print debugging info */
     if (debugging)
     {
+        fprintf( outfile, "\tpushl $%d\n", reg_func );
+
         if ( UsePIC )
             fprintf( outfile, "\tcall " PREFIX "RELAY_DebugCallTo16Ret@PLT\n" );
         else
             fprintf( outfile, "\tcall " PREFIX "RELAY_DebugCallTo16Ret\n" );
+
+        fprintf( outfile, "\taddl $4, %%esp\n" );
     }
 
     /* Leave Win16 Mutex */
@@ -657,17 +685,20 @@ static void BuildRet16Func( FILE *outfile )
     fprintf( outfile, "\t.globl " PREFIX "CallTo16_Ret\n" );
     fprintf( outfile, PREFIX "CallTo16_Ret:\n" );
 
+    /* Save %esp into %esi */
+    fprintf( outfile, "\tmovl %%esp,%%esi\n" );
+
     /* Restore 32-bit segment registers */
 
-    fprintf( outfile, "\tmovw $0x%04x,%%bx\n", data_selector );
+    fprintf( outfile, "\tmovw $0x%04x,%%di\n", data_selector );
 #ifdef __svr4__
     fprintf( outfile, "\tdata16\n");
 #endif
-    fprintf( outfile, "\tmovw %%bx,%%ds\n" );
+    fprintf( outfile, "\tmovw %%di,%%ds\n" );
 #ifdef __svr4__
     fprintf( outfile, "\tdata16\n");
 #endif
-    fprintf( outfile, "\tmovw %%bx,%%es\n" );
+    fprintf( outfile, "\tmovw %%di,%%es\n" );
 
     fprintf( outfile, "\tmovw " PREFIX "SYSLEVEL_Win16CurrentTeb,%%fs\n" );
 
@@ -676,7 +707,7 @@ static void BuildRet16Func( FILE *outfile )
 #ifdef __svr4__
     fprintf( outfile, "\tdata16\n");
 #endif
-    fprintf( outfile, "\tmovw %%bx,%%ss\n" );
+    fprintf( outfile, "\tmovw %%di,%%ss\n" );
     fprintf( outfile, "\t.byte 0x64\n\tmovl (%d),%%esp\n", STACKOFFSET );
     fprintf( outfile, "\t.byte 0x64\n\tpopl (%d)\n", STACKOFFSET );
 
diff --git a/windows/winproc.c b/windows/winproc.c
index 5afea862b4a..87172379b9c 100644
--- a/windows/winproc.c
+++ b/windows/winproc.c
@@ -238,7 +238,8 @@ static LRESULT WINAPI WINPROC_CallWndProc16( WNDPROC16 proc, HWND16 hwnd,
     args[3] = msg;
     args[4] = hwnd;
 
-    ret = CallTo16RegisterShort( &context, 5 * sizeof(WORD) );
+    CallTo16RegisterShort( &context, 5 * sizeof(WORD) );
+    ret = MAKELONG( AX_reg(&context), DX_reg(&context) );
     if (offset) stack16_pop( offset );
 
     WIN_RestoreWndsLock(iWndsLocks);