diff --git a/TODO b/TODO
index 647d33f3..dd08db4b 100644
--- a/TODO
+++ b/TODO
@@ -1,5 +1,3 @@
-- bicubic needs a signed int lut path with rounding in the other direction
-  for negatives
 
 - vips_object_unref_outputs() needs docs ... bindings will need it
  
diff --git a/libvips/resample/bicubic.cpp b/libvips/resample/bicubic.cpp
index 66bfa5b6..238bf991 100644
--- a/libvips/resample/bicubic.cpp
+++ b/libvips/resample/bicubic.cpp
@@ -104,9 +104,10 @@ G_DEFINE_TYPE( VipsInterpolateBicubic, vips_interpolate_bicubic,
 
 /* Fixed-point version, for 8 and 16-bit types.
  */
-template <typename T, int min_value, int max_value>
+
+template <typename T, int max_value>
 static void inline
-bicubic_int_tab( void *pout, const VipsPel *pin,
+bicubic_unsigned_int_tab( void *pout, const VipsPel *pin,
 	const int bands, const int lskip,
 	const int *cx, const int *cy )
 {
@@ -152,7 +153,73 @@ bicubic_int_tab( void *pout, const VipsPel *pin,
 		const T qua_thr = in[l3_plus_b2];
 		const T qua_fou = in[l3_plus_b3];
 
-		int bicubic = bicubic_int<T>(
+		int bicubic = bicubic_unsigned_int<T>(
+			uno_one, uno_two, uno_thr, uno_fou,
+			dos_one, dos_two, dos_thr, dos_fou,
+			tre_one, tre_two, tre_thr, tre_fou,
+			qua_one, qua_two, qua_thr, qua_fou,
+			cx, cy );
+
+		if( bicubic < 0 )
+			bicubic = 0;
+		else if( bicubic > max_value )
+			bicubic = max_value;
+
+		out[z] = bicubic;
+
+		in += 1;
+	}
+}
+
+template <typename T, int min_value, int max_value>
+static void inline
+bicubic_signed_int_tab( void *pout, const VipsPel *pin,
+	const int bands, const int lskip,
+	const int *cx, const int *cy )
+{
+	T* restrict out = (T *) pout;
+	const T* restrict in = (T *) pin;
+
+	const int b1 = bands;
+	const int b2 = b1 + b1;
+	const int b3 = b1 + b2;
+
+	const int l1 = lskip / sizeof( T );
+	const int l2 = l1 + l1;
+	const int l3 = l1 + l2;
+
+        const int l1_plus_b1 = l1 + b1;
+        const int l1_plus_b2 = l1 + b2;
+        const int l1_plus_b3 = l1 + b3;
+        const int l2_plus_b1 = l2 + b1;
+        const int l2_plus_b2 = l2 + b2;
+        const int l2_plus_b3 = l2 + b3;
+        const int l3_plus_b1 = l3 + b1;
+        const int l3_plus_b2 = l3 + b2;
+        const int l3_plus_b3 = l3 + b3;
+
+	for( int z = 0; z < bands; z++ ) {
+		const T uno_one = in[0];
+		const T uno_two = in[b1];
+		const T uno_thr = in[b2];
+		const T uno_fou = in[b3];
+
+		const T dos_one = in[l1];
+		const T dos_two = in[l1_plus_b1];
+		const T dos_thr = in[l1_plus_b2];
+		const T dos_fou = in[l1_plus_b3];
+
+		const T tre_one = in[l2];
+		const T tre_two = in[l2_plus_b1];
+		const T tre_thr = in[l2_plus_b2];
+		const T tre_fou = in[l2_plus_b3];
+
+		const T qua_one = in[l3];
+		const T qua_two = in[l3_plus_b1];
+		const T qua_thr = in[l3_plus_b2];
+		const T qua_fou = in[l3_plus_b3];
+
+		int bicubic = bicubic_signed_int<T>(
 			uno_one, uno_two, uno_thr, uno_fou,
 			dos_one, dos_two, dos_thr, dos_fou,
 			tre_one, tre_two, tre_thr, tre_fou,
@@ -350,7 +417,7 @@ vips_interpolate_bicubic_interpolate( VipsInterpolate *interpolate,
 
 	switch( in->im->BandFmt ) {
 	case VIPS_FORMAT_UCHAR:
-		bicubic_int_tab<unsigned char, 0, UCHAR_MAX>(
+		bicubic_unsigned_int_tab<unsigned char, UCHAR_MAX>(
 			out, p, bands, lskip,
 			cxi, cyi );
 
@@ -371,19 +438,19 @@ vips_interpolate_bicubic_interpolate( VipsInterpolate *interpolate,
 		break;
 
 	case VIPS_FORMAT_CHAR:
-		bicubic_int_tab<signed char, SCHAR_MIN, SCHAR_MAX>(
+		bicubic_signed_int_tab<signed char, SCHAR_MIN, SCHAR_MAX>(
 			out, p, bands, lskip,
 			cxi, cyi );
 		break;
 
 	case VIPS_FORMAT_USHORT:
-		bicubic_int_tab<unsigned short, 0, USHRT_MAX>(
+		bicubic_unsigned_int_tab<unsigned short, USHRT_MAX>(
 			out, p, bands, lskip,
 			cxi, cyi );
 		break;
 
 	case VIPS_FORMAT_SHORT:
-		bicubic_int_tab<signed short, SHRT_MIN, SHRT_MAX>(
+		bicubic_signed_int_tab<signed short, SHRT_MIN, SHRT_MAX>(
 			out, p, bands, lskip,
 			cxi, cyi );
 		break;
diff --git a/libvips/resample/templates.h b/libvips/resample/templates.h
index 7cb0e99a..4ac9027c 100644
--- a/libvips/resample/templates.h
+++ b/libvips/resample/templates.h
@@ -169,49 +169,103 @@ bilinear_nosign(
  * Bicubic (Catmull-Rom) interpolation templates:
  */
 
+static int inline
+unsigned_fixed_round( int v )
+{
+	const int round_by = VIPS_INTERPOLATE_SCALE >> 1;
+
+	return( (v + round_by) >> VIPS_INTERPOLATE_SHIFT );
+}
+
 /* Fixed-point integer bicubic, used for 8 and 16-bit types.
  */
 template <typename T> static int inline
-bicubic_int(
+bicubic_unsigned_int(
 	const T uno_one, const T uno_two, const T uno_thr, const T uno_fou,
 	const T dos_one, const T dos_two, const T dos_thr, const T dos_fou,
 	const T tre_one, const T tre_two, const T tre_thr, const T tre_fou,
 	const T qua_one, const T qua_two, const T qua_thr, const T qua_fou,
 	const int* restrict cx, const int* restrict cy )
 {
-	const int r0 =
-		(cx[0] * uno_one +
-		 cx[1] * uno_two +
-		 cx[2] * uno_thr +
-		 cx[3] * uno_fou +
-		 (VIPS_INTERPOLATE_SCALE >> 1)) >> VIPS_INTERPOLATE_SHIFT;
+	const int r0 = unsigned_fixed_round( 
+		cx[0] * uno_one +
+		cx[1] * uno_two +
+		cx[2] * uno_thr +
+		cx[3] * uno_fou ); 
 
-	const int r1 =
-		(cx[0] * dos_one +
-		 cx[1] * dos_two +
-		 cx[2] * dos_thr +
-		 cx[3] * dos_fou +
-		 (VIPS_INTERPOLATE_SCALE >> 1)) >> VIPS_INTERPOLATE_SHIFT;
+	const int r1 = unsigned_fixed_round( 
+		cx[0] * dos_one +
+		cx[1] * dos_two +
+		cx[2] * dos_thr +
+		cx[3] * dos_fou );
 
-	const int r2 =
-		(cx[0] * tre_one +
-		 cx[1] * tre_two +
-		 cx[2] * tre_thr +
-		 cx[3] * tre_fou +
-		 (VIPS_INTERPOLATE_SCALE >> 1)) >> VIPS_INTERPOLATE_SHIFT;
+	const int r2 = unsigned_fixed_round( 
+		cx[0] * tre_one +
+		cx[1] * tre_two +
+		cx[2] * tre_thr +
+		cx[3] * tre_fou );
 
-	const int r3 =
-		(cx[0] * qua_one +
-		 cx[1] * qua_two +
-		 cx[2] * qua_thr +
-		 cx[3] * qua_fou +
-		 (VIPS_INTERPOLATE_SCALE >> 1)) >> VIPS_INTERPOLATE_SHIFT;
+	const int r3 = unsigned_fixed_round( 
+		cx[0] * qua_one +
+		cx[1] * qua_two +
+		cx[2] * qua_thr +
+		cx[3] * qua_fou );
 
-	return( (cy[0] * r0 +
-		 cy[1] * r1 +
-		 cy[2] * r2 +
-		 cy[3] * r3 +
-		 (VIPS_INTERPOLATE_SCALE >> 1)) >> VIPS_INTERPOLATE_SHIFT );
+	return( unsigned_fixed_round( 
+		cy[0] * r0 +
+		cy[1] * r1 +
+		cy[2] * r2 +
+		cy[3] * r3 ) );
+}
+
+static int inline
+signed_fixed_round( int v )
+{
+	const int sign_of_v = 2 * (v > 0) - 1;
+	const int round_by = sign_of_v * (VIPS_INTERPOLATE_SCALE >> 1);
+
+	return( (v + round_by) >> VIPS_INTERPOLATE_SHIFT );
+}
+
+/* Fixed-point integer bicubic, used for 8 and 16-bit types.
+ */
+template <typename T> static int inline
+bicubic_signed_int(
+	const T uno_one, const T uno_two, const T uno_thr, const T uno_fou,
+	const T dos_one, const T dos_two, const T dos_thr, const T dos_fou,
+	const T tre_one, const T tre_two, const T tre_thr, const T tre_fou,
+	const T qua_one, const T qua_two, const T qua_thr, const T qua_fou,
+	const int* restrict cx, const int* restrict cy )
+{
+	const int r0 = signed_fixed_round( 
+		cx[0] * uno_one +
+		cx[1] * uno_two +
+		cx[2] * uno_thr +
+		cx[3] * uno_fou ); 
+
+	const int r1 = signed_fixed_round( 
+		cx[0] * dos_one +
+		cx[1] * dos_two +
+		cx[2] * dos_thr +
+		cx[3] * dos_fou );
+
+	const int r2 = signed_fixed_round( 
+		cx[0] * tre_one +
+		cx[1] * tre_two +
+		cx[2] * tre_thr +
+		cx[3] * tre_fou );
+
+	const int r3 = signed_fixed_round( 
+		cx[0] * qua_one +
+		cx[1] * qua_two +
+		cx[2] * qua_thr +
+		cx[3] * qua_fou );
+
+	return( signed_fixed_round( 
+		cy[0] * r0 +
+		cy[1] * r1 +
+		cy[2] * r2 +
+		cy[3] * r3 ) );
 }
 
 /* Floating-point bicubic, used for int/float/double types.