diff --git a/TODO.txt b/TODO.txt
index 9bc05571..d62c1c09 100644
--- a/TODO.txt
+++ b/TODO.txt
@@ -8,7 +8,8 @@ Version 0.10
   depending on which optional dependencies are available.
 * Remove deprecated `out` parameter of `skimage.morphology.binary_*`
 * Remove deprecated parameter `depth` in `skimage.segmentation.random_walker`
-* Remove deprecated logger function in ``skimage/__init__.py``
+* Remove deprecated logger function in `skimage/__init__.py`
+* Remove deprecated function `filter.median_filter`
 
 Version 0.9
 -----------
diff --git a/skimage/filter/ctmf.py b/skimage/filter/ctmf.py
index 267c5b6c..e39aa8bc 100644
--- a/skimage/filter/ctmf.py
+++ b/skimage/filter/ctmf.py
@@ -1,4 +1,4 @@
-'''ctmf.py - constant time per pixel median filtering with an octagonal shape
+"""ctmf.py - constant time per pixel median filtering with an octagonal shape
 
 Reference: S. Perreault and P. Hebert, "Median Filtering in Constant Time",
 IEEE Transactions on Image Processing, September 2007.
@@ -9,39 +9,46 @@ Copyright (c) 2003-2009 Massachusetts Institute of Technology
 Copyright (c) 2009-2011 Broad Institute
 All rights reserved.
 Original author: Lee Kamentsky
-'''
+"""
 
+import warnings
 import numpy as np
 from . import _ctmf
 from ._rank_order import rank_order
+from .._shared.utils import deprecated
 
 
+@deprecated('filter.rank.median')
 def median_filter(image, radius=2, mask=None, percent=50):
-    '''Masked median filter with octagon shape.
+    """Masked median filter with octagon shape.
 
     Parameters
     ----------
-    image : (M,N) ndarray, dtype uint8
+    image : (M, N) ndarray
         Input image.
-    radius : {int, 2}, optional
-        The radius of a circle inscribed into the filtering
-        octagon. Must be at least 2.  Default radius is 2.
-    mask : (M,N) ndarray, dtype uint8, optional
-        A value of 1 indicates a significant pixel, 0
-        that a pixel is masked.  By default, all pixels
-        are considered.
-    percent : {int, 50}, optional
+    radius : int
+        Radius (in pixels) of a circle inscribed into the filtering
+        octagon. Must be at least 2. Default radius is 2.
+    mask : (M, N) ndarray
+        Mask with 1's for significant pixels, 0's for masked pixels.
+        By default, all pixels are considered significant.
+    percent : int
         The unmasked pixels within the octagon are sorted, and the
-        value at the `percent`-th index chosen.  For example, the
-        default value of 50 chooses the median pixel.
+        value at `percent` percent of the index range is chosen.
+        Default value of 50 gives the median pixel.
 
     Returns
     -------
-    out : (M,N) ndarray, dtype uint8
-        Filtered array.  In areas where the median filter does
-        not overlap the mask, the filtered result is underfined, but
+    out : (M, N) ndarray
+        Filtered array. In areas where the median filter does
+        not overlap the mask, the filtered result is undefined, but
         in practice, it will be the lowest value in the valid area.
 
+    Notes
+    -----
+    Because of the histogram implementation, the number of unique values
+    for the output is limited to 256.
+
     Examples
     --------
     >>> a = np.ones((5, 5))
@@ -49,53 +56,54 @@ def median_filter(image, radius=2, mask=None, percent=50):
     >>> b = median_filter(a)
     >>> b[2, 2] # the median filter is good at removing outliers
     1.0
-    '''
+    """
 
     if image.ndim != 2:
-        raise TypeError("The input 'image' must be a two dimensional array.")
+        raise TypeError("Input 'image' must be a two-dimensional array.")
 
     if radius < 2:
-        raise ValueError("The input 'radius' must be >= 2.")
+        raise ValueError("Input 'radius' must be >= 2.")
 
     if mask is None:
         mask = np.ones(image.shape, dtype=np.bool)
     mask = np.ascontiguousarray(mask, dtype=np.bool)
 
     if np.all(~ mask):
+        warnings.warn('Mask is all over image! Returning copy of input image.')
         return image.copy()
-    #
-    # Normalize the ranked image to 0-255
-    #
+    
     if (not np.issubdtype(image.dtype, np.int) or
         np.min(image) < 0 or np.max(image) > 255):
-        ranked_image, translation = rank_order(image[mask])
-        max_ranked_image = np.max(ranked_image)
-        if max_ranked_image == 0:
-            return image
-        if max_ranked_image > 255:
-            ranked_image = ranked_image * 255 // max_ranked_image
+        ranked_values, translation = rank_order(image[mask])
+        max_ranked_values = np.max(ranked_values)
+        if max_ranked_values == 0:
+            warnings.warn('Particular case? Returning copy of input image.')
+            return image.copy()
+        if max_ranked_values > 255:
+            ranked_values = ranked_values * 255 // max_ranked_values
         was_ranked = True
     else:
-        ranked_image = image[mask]
+        ranked_values = image[mask]
         was_ranked = False
-    input = np.zeros(image.shape, np.uint8)
-    input[mask] = ranked_image
+    ranked_image = np.zeros(image.shape, np.uint8)
+    ranked_image[mask] = ranked_values
 
     mask.dtype = np.uint8
     output = np.zeros(image.shape, np.uint8)
 
-    _ctmf.median_filter(input, mask, output, radius, percent)
+    _ctmf.median_filter(ranked_image, mask, output, radius, percent)
     if was_ranked:
         #
         # The translation gives the original value at each ranking.
         # We rescale the output to the original ranking and then
         # use the translation to look up the original value in the image.
         #
-        if max_ranked_image > 255:
+        if max_ranked_values > 255:
             result = translation[output.astype(np.uint32) *
-                                 max_ranked_image // 255]
+                                 max_ranked_values // 255]
         else:
             result = translation[output]
     else:
         result = output
     return result
+