diff --git a/skimage/filter/tests/test_tv_denoise.py b/skimage/filter/tests/test_tv_denoise.py
index 9869d992..f851f4f1 100644
--- a/skimage/filter/tests/test_tv_denoise.py
+++ b/skimage/filter/tests/test_tv_denoise.py
@@ -15,6 +15,8 @@ class TestTvDenoise():
         lena = color.rgb2gray(data.lena())[:256, :256]
         # add noise to lena
         lena += 0.5 * lena.std()*np.random.randn(*lena.shape)
+        # clip noise so that it does not exceed allowed range for float images.
+        lena = np.clip(lena, 0, 1)
         # denoise
         denoised_lena = filter.tv_denoise(lena, weight=60.0)
         # which dtype?
diff --git a/skimage/util/dtype.py b/skimage/util/dtype.py
index da2ff5e7..ac8f305b 100644
--- a/skimage/util/dtype.py
+++ b/skimage/util/dtype.py
@@ -20,7 +20,7 @@ _supported_types = (np.uint8, np.uint16, np.uint32,
                     np.float16, np.float32, np.float64)
 
 
-def _convert(image, dtype):
+def convert(image, dtype):
     """
     Convert an image to the requested data-type.
 
@@ -63,6 +63,8 @@ def _convert(image, dtype):
                  "%s to %s" % (dtypeobj_in, dtypeobj))
 
     if kind_in == 'f':
+        if np.min(image) < 0 or np.max(image) > 1:
+            raise ValueError("Images of type float must be between 0 and 1")
         if kind == 'f':
             # floating point -> floating point
             if itemsize_in > itemsize:
@@ -120,7 +122,7 @@ def _convert(image, dtype):
             # int8 -> uint32
             image = dtype(image)
             image *= 2**16 + 2**8 + 1
-        result += image
+        result += dtype(image)
         return result
     if kind == 'i':
         # signed integer -> signed integer
@@ -165,7 +167,7 @@ def img_as_float(image):
     Negative input values will be shifted to the positive domain.
 
     """
-    return _convert(image, np.float64)
+    return convert(image, np.float64)
 
 
 def img_as_uint(image):
@@ -186,7 +188,7 @@ def img_as_uint(image):
     Negative input values will be shifted to the positive domain.
 
     """
-    return _convert(image, np.uint16)
+    return convert(image, np.uint16)
 
 
 def img_as_int(image):
@@ -208,7 +210,7 @@ def img_as_int(image):
     the output image will still only have positive values.
 
     """
-    return _convert(image, np.int16)
+    return convert(image, np.int16)
 
 
 def img_as_ubyte(image):
@@ -230,4 +232,4 @@ def img_as_ubyte(image):
     the output image will still only have positive values.
 
     """
-    return _convert(image, np.uint8)
+    return convert(image, np.uint8)
diff --git a/skimage/util/tests/test_dtype.py b/skimage/util/tests/test_dtype.py
index 4932a480..11dd7248 100644
--- a/skimage/util/tests/test_dtype.py
+++ b/skimage/util/tests/test_dtype.py
@@ -1,7 +1,9 @@
 import numpy as np
-from numpy.testing import assert_equal
+from numpy.testing import assert_equal, assert_raises
 from skimage import img_as_int, img_as_float, \
                     img_as_uint, img_as_ubyte
+from skimage.util.dtype import convert
+
 
 dtype_range = {np.uint8: (0, 255),
                np.uint16: (0, 65535),
@@ -34,6 +36,44 @@ def test_range():
                   "From %s to %s" % (np.dtype(dtype), np.dtype(dt)), \
                   y, omin, omax
 
+
+def test_range_extra_dtypes():
+    """Test code paths that are not skipped by `test_range`"""
+
+    # Add non-standard data types that are allowed by the `convert` function.
+    dtype_range_extra = dtype_range.copy()
+    dtype_range_extra.update({np.int32: (-2147483648, 2147483647),
+                              np.uint32: (0, 4294967295)})
+
+    dtype_pairs = [(np.uint8, np.uint32),
+                   (np.int8, np.uint32),
+                   (np.int8, np.int32),
+                   (np.int32, np.int8),
+                   (np.float64, np.float32),
+                   (np.int32, np.float32)]
+
+    for dtype_in, dt in dtype_pairs:
+        imin, imax = dtype_range_extra[dtype_in]
+        x = np.linspace(imin, imax, 10).astype(dtype_in)
+        y = convert(x, dt)
+        omin, omax = dtype_range_extra[dt]
+        yield _verify_range, \
+              "From %s to %s" % (np.dtype(dtype_in), np.dtype(dt)), \
+              y, omin, omax
+
+
+def test_unsupported_dtype():
+    x = np.arange(10).astype(np.uint64)
+    assert_raises(ValueError, img_as_int, x)
+
+
+def test_float_out_of_range():
+    too_high = np.array([2], dtype=np.float32)
+    assert_raises(ValueError, img_as_int, too_high)
+    too_low = np.array([-1], dtype=np.float32)
+    assert_raises(ValueError, img_as_int, too_low)
+
+
 if __name__ == '__main__':
     np.testing.run_module_suite()