diff --git a/ChangeLog b/ChangeLog
index 753a7cfe..745496df 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -6,6 +6,8 @@
 - im_histplot() could fail for signed int histograms
 - im_fwfft() and im_invfft() could free their output image too early
 - added im_local_imask(), im_local_dmask()
+- added im_mpercent_hist()
+- im_maplut() casts the index image to one of the uint types
 
 16/1/10 started 7.21.2
 - "invalidate" is careful to keep images alive, so invalidate callbacks can do
diff --git a/TODO b/TODO
index 5f459b09..2f178fd8 100644
--- a/TODO
+++ b/TODO
@@ -1,4 +1,9 @@
 
+- expose more of the tone funcs in nip2
+
+- quite a few hist operations have no GUI ... lhisteq, for example? or
+  histspec?
+
 - conv with scale == 0 gives /0 error
 
 - added im_local_imask(), im_local_dmask(), needs docs?
@@ -7,22 +12,20 @@
 
   is local_imask() in the best place? shouldn't it be in mask.h?
 
-- im_tone_analyse() needs docs and testing
-
-  also im_tone_map() 
-
-- im_rotate_imask90 only works for square, odd-sized massks, argh
+- im_rotate_imask90 only works for square, odd-sized masks, argh
 
 - lots of stupid little files in hist, eg. im_hsp.c ... paste them into larger 
   modules
 
-- quite a few hist operations have no GUI ... lhisteq, for example? or
-  histspec?
-
 - check the order of decls in hist.h, is it sensible?
 
   check the order in freq_flt as well
 
+
+
+
+
+
 - im_render() should be renamed as vips_sink_screen()? also vips_sink_disc()
   instead of vips_discsink()?
 
diff --git a/libvips/histograms_lut/im_maplut.c b/libvips/histograms_lut/im_maplut.c
index fcaa4549..f0e43a9e 100644
--- a/libvips/histograms_lut/im_maplut.c
+++ b/libvips/histograms_lut/im_maplut.c
@@ -534,6 +534,19 @@ maplut_gen( REGION *or, void *vseq, void *a, void *b )
 	return( 0 );
 }
 
+/* Save a bit of typing.
+ */
+#define UC IM_BANDFMT_UCHAR
+#define US IM_BANDFMT_USHORT
+#define UI IM_BANDFMT_UINT
+
+/* Type mapping: go to uchar or ushort.
+ */
+static int bandfmt_maplut[10] = {
+/* UC   C  US   S  UI   I   F   X  D   DX */
+   UC, UC, US, US, UI, UI, UI, UI, UI, UI
+};
+
 /**
  * im_maplut:
  * @in: input image
@@ -543,8 +556,8 @@ maplut_gen( REGION *or, void *vseq, void *a, void *b )
  * Map an image through another image acting as a LUT (Look Up Table). 
  * The lut may have any type, and the output image will be that type.
  * 
- * The input image must be an unsigned integer types, that is, it must
- * be one of IM_BANDFMT_UCHAR, IM_BANDFMT_USHORT or IM_BANDFMT_UINT.
+ * The input image will be cast to one of the unsigned integer types, that is,
+ * IM_BANDFMT_UCHAR, IM_BANDFMT_USHORT or IM_BANDFMT_UINT.
  * 
  * If @lut is too small for the input type (for example, if @in is
  * IM_BANDFMT_UCHAR but @lut only has 100 elements), the lut is padded out
@@ -564,6 +577,7 @@ maplut_gen( REGION *or, void *vseq, void *a, void *b )
 int 
 im_maplut( IMAGE *in, IMAGE *out, IMAGE *lut )
 {
+	IMAGE *t;
 	LutInfo *st;
 
 	/* Check input output. Old-style IO from lut, for simplicity.
@@ -572,14 +586,19 @@ im_maplut( IMAGE *in, IMAGE *out, IMAGE *lut )
 		im_check_uncoded( "im_maplut", lut ) ||
 		im_check_uncoded( "im_maplut", in ) ||
 		im_check_bands_1orn( "im_maplut", in, lut ) ||
-		im_check_uint( "im_maplut", in ) ||
 		im_piocheck( in, out ) || 
 		im_incheck( lut ) )
 		return( -1 );
 
+	/* Cast in to u8/u16.
+	 */
+	if( !(t = im_open_local( out, "im_maplut", "p" )) ||
+		im_clip2fmt( in, t, bandfmt_maplut[in->BandFmt] ) )
+		return( -1 );
+
 	/* Prepare the output header.
 	 */
-        if( im_cp_descv( out, in, lut, NULL ) )
+        if( im_cp_descv( out, t, lut, NULL ) )
                 return( -1 );
 
 	/* Force output to be the same type as lut.
@@ -599,12 +618,12 @@ im_maplut( IMAGE *in, IMAGE *out, IMAGE *lut )
 
 	/* Set demand hints.
 	 */
-	if( im_demand_hint( out, IM_THINSTRIP, in, NULL ) )
+	if( im_demand_hint( out, IM_THINSTRIP, t, NULL ) )
 		return( -1 );
 
 	/* Process!
 	 */
-        if( im_generate( out, maplut_start, maplut_gen, maplut_stop, in, st ) )
+        if( im_generate( out, maplut_start, maplut_gen, maplut_stop, t, st ) )
                 return( -1 );
 
         return( 0 );
diff --git a/libvips/histograms_lut/im_mpercent.c b/libvips/histograms_lut/im_mpercent.c
index 2f9bad7a..4f133d8d 100644
--- a/libvips/histograms_lut/im_mpercent.c
+++ b/libvips/histograms_lut/im_mpercent.c
@@ -53,6 +53,53 @@
 #include <dmalloc.h>
 #endif /*WITH_DMALLOC*/
 
+/**
+ * im_mpercent_hist:
+ * @hist: input histogram image
+ * @percent: threshold percentage
+ * @out: output threshold value
+ *
+ * Just like im_mpercent(), except it works on an image histogram. Handy if
+ * you want to run im_mpercent() several times without having to recompute the
+ * histogram each time.
+ *
+ * See also: im_mpercent().
+ *
+ * Returns: 0 on success, -1 on error
+ */
+int
+im_mpercent_hist( IMAGE *hist, double percent, int *out )
+{
+	IMAGE *base;
+	IMAGE *t[6];
+	double pos;
+
+	if( im_check_hist( "im_mpercent", hist ) )
+		return( -1 );
+
+	if( !(base = im_open( "im_mpercent", "p" )) )
+		return( -1 );
+	if( im_open_local_array( base, t, 6, "im_mpercent", "p" ) ) {
+		im_close( base );
+		return( -1 );
+	}
+
+	if( im_histcum( hist, t[1] ) ||
+		im_histnorm( t[1], t[2] ) ||
+		im_lessconst( t[2], t[3], percent * t[2]->Xsize ) ||
+		im_fliphor( t[3], t[4] ) ||
+		im_profile( t[4], t[5], 1 ) ||
+		im_avg( t[5], &pos ) ) {
+		im_close( base );
+		return( -1 );
+	}
+	im_close( base );
+
+	*out = pos;
+
+	return( 0 );
+}
+
 /**
  * im_mpercent:
  * @in: input image
@@ -73,31 +120,17 @@
  */
 int
 im_mpercent( IMAGE *in, double percent, int *out )
-{	
-	IMAGE *base;
-	IMAGE *t[6];
-	double pos;
+{
+	IMAGE *t;
 
-	if( !(base = im_open( "im_mpercent1", "p" )) )
+	if( !(t = im_open( "im_mpercent1", "p" )) )
 		return( -1 );
-	if( im_open_local_array( base, t, 6, "im_mpercent", "p" ) ) {
-		im_close( base );
+	if( im_histgr( in, t, -1 ) ||
+		im_mpercent_hist( t, percent, out ) ) {
+		im_close( t );
 		return( -1 );
 	}
-
-	if( im_histgr( in, t[0], -1 ) ||
-		im_histcum( t[0], t[1] ) ||
-		im_histnorm( t[1], t[2] ) ||
-		im_lessconst( t[2], t[3], percent * t[2]->Xsize ) ||
-		im_fliphor( t[3], t[4] ) ||
-		im_profile( t[4], t[5], 1 ) ||
-		im_avg( t[5], &pos ) ) {
-		im_close( base );
-		return( -1 );
-	}
-	im_close( base );
-
-	*out = pos;
+	im_close( t );
 
 	return( 0 );
 }
diff --git a/libvips/histograms_lut/tone.c b/libvips/histograms_lut/tone.c
index 40c51809..b3f7bd72 100644
--- a/libvips/histograms_lut/tone.c
+++ b/libvips/histograms_lut/tone.c
@@ -427,9 +427,23 @@ im_tone_map( IMAGE *in, IMAGE *out, IMAGE *lut )
 	return( im_copy( t[4], out ) );
 }
 
-/* Find histogram of in, and use that to set Lb, Lw levels.
- * @(#) im_tone_analyse: find the histogram of a LabS or LabQ image and use
- * @(#) that to set the Ln and Lw parameters of im_tone_build()
+/**
+ * im_tone_analyse:
+ * in: input image 
+ * out: output image 
+ * Ps: shadow point (eg. 0.2)
+ * Pm: mid-tone point (eg. 0.5)
+ * Ph: highlight point (eg. 0.8)
+ * S: shadow adjustment (+/- 30)
+ * M: mid-tone adjustment (+/- 30)
+ * H: highlight adjustment (+/- 30)
+ *
+ * As im_tone_build(), but analyse the histogram of @in and use it to
+ * pick the 0.1% and 99.9% points for @Lb and @Lw.
+ *
+ * See also: im_tone_build().
+ *
+ * Returns: 0 on success, -1 on error
  */
 int 
 im_tone_analyse( 
@@ -466,8 +480,8 @@ im_tone_analyse(
 		im_histgr( t[2], t[3], -1 ) )
 		return( -1 );
 
-	if( im_mpercent( t[3], 0.1 / 100.0, &low ) ||
-		im_mpercent( t[3], 99.9 / 100.0, &high ) )
+	if( im_mpercent_hist( t[3], 0.1 / 100.0, &high ) ||
+		im_mpercent_hist( t[3], 99.9 / 100.0, &low ) )
 		return( -1 );
 
 	Lb = 100 * low / 32768;
diff --git a/libvips/include/vips/histograms_lut.h b/libvips/include/vips/histograms_lut.h
index 853e1cb0..9ca8b8e8 100644
--- a/libvips/include/vips/histograms_lut.h
+++ b/libvips/include/vips/histograms_lut.h
@@ -70,6 +70,7 @@ int im_ismonotonic( IMAGE *lut, int *out );
 int im_tone_map( IMAGE *in, IMAGE *out, IMAGE *lut );
 int im_project( IMAGE *in, IMAGE *hout, IMAGE *vout );
 int im_mpercent( IMAGE *in, double percent, int *out );
+int im_mpercent_hist( IMAGE *hist, double percent, int *out );
 
 #ifdef __cplusplus
 }