diff --git a/TODO b/TODO
index 0475cf41..deae2057 100644
--- a/TODO
+++ b/TODO
@@ -1,4 +1,27 @@
 
+- maxpos_avg seems to give variable results
+
+	irb(main):022:0> im.maxpos_avg
+	=> [141.0, 138.0, 248.0]
+	irb(main):023:0> im.maxpos_avg
+	=> [73.0, 95.0, 248.0]
+
+  [107.0, 116.5, 248.0] is correct, checked by hand
+
+$ vips max --size 10 spec/samples/wagon.v --x-array --y-array --out-array
+72 107 78 73 74 75 99 100 141 73 
+95 141 94 94 94 95 131 131 138 95 
+245 246 246 247 247 247 247 247 248 248 
+248.000000
+
+  so there are two maxes of 248, result should be average of the two
+
+  seems to be a race in maxpos_avg, it's still some Tom code, replace with a
+  wrapper over vips_max()
+
+
+
+
 - threadpool recyling might be useful for dzsave?
 
 
diff --git a/libvips/conversion/scale.c b/libvips/conversion/scale.c
index 181e9aa1..e08197bb 100644
--- a/libvips/conversion/scale.c
+++ b/libvips/conversion/scale.c
@@ -98,6 +98,7 @@ vips_scale_build( VipsObject *object )
 		/* Range of zero: just return black.
 		 */
 		if( vips_black( &t[1], scale->in->Xsize, scale->in->Ysize, 
+			"bands", scale->in->Bands,
 			NULL ) ||
 			vips_image_write( t[1], conversion->out ) )
 			return( -1 );