diff --git a/ivl_target.h b/ivl_target.h
index 33b2adb5f..bc5498a52 100644
--- a/ivl_target.h
+++ b/ivl_target.h
@@ -992,8 +992,9 @@ extern unsigned ivl_lpm_lineno(ivl_lpm_t net);
  * more than the width of the output, although the possibility of
  * overflow exists at run time.
  *
- * Multiply may be signed. If so, the output should be sign extended
- * to fill in its result.
+ * The inputs are always treated as unsigned. If the expression is
+ * supposed to be signed, elaboration will generate the necessary sign
+ * extension, so the target need not (must not) consider signedness.
  *
  * - Power (IVL_LPM_POW)
  * The power takes two inputs and generates an output. Unlike other
diff --git a/vvp/arith.cc b/vvp/arith.cc
index 9a2061504..126cfdaa1 100644
--- a/vvp/arith.cc
+++ b/vvp/arith.cc
@@ -336,13 +336,13 @@ void vvp_arith_mult::recv_vec4(vvp_net_ptr_t ptr, const vvp_vector4_t&bit)
       }
 
       long a;
-      if (! vector4_to_value(op_a_, a, true, true)) {
+      if (! vector4_to_value(op_a_, a, false, true)) {
 	    vvp_send_vec4(ptr.ptr()->out, x_val_);
 	    return;
       }
 
       long b;
-      if (! vector4_to_value(op_b_, b, true, true)) {
+      if (! vector4_to_value(op_b_, b, false, true)) {
 	    vvp_send_vec4(ptr.ptr()->out, x_val_);
 	    return;
       }