diff --git a/ComputeNormalMap.md b/ComputeNormalMap.md
new file mode 100644
index 0000000..dabd373
--- /dev/null
+++ b/ComputeNormalMap.md
@@ -0,0 +1,53 @@
+Converts a height-map to a normal-map.
+
+    HRESULT ComputeNormalMap( const Image& srcImage, DWORD flags,
+        float amplitude, DXGI_FORMAT format,
+        ScratchImage& normalMap );
+
+    HRESULT ComputeNormalMap( const Image* srcImages, size_t nimages,
+        const TexMetadata& metadata,
+        DWORD flags, float amplitude, DXGI_FORMAT format,
+        ScratchImage& normalMaps );
+
+# Parameters
+
+_flags_: a combination of the following flags
+* ``CNMAP_DEFAULT`` Default flags
+
+Selects which channel to use as the height. Luminance is the average of R, G, and B.
+* ``CNMAP_CHANNEL_RED``
+* ``CNMAP_CHANNEL_GREEN``
+* ``CNMAP_CHANNEL_BLUE``
+* ``CNMAP_CHANNEL_ALPHA``
+* ``CNMAP_CHANNEL_LUMINANCE``
+
+Selects mirroring semantics for scanline references. Otherwise defaults to wrap.
+* ``CNMAP_MIRROR_U``
+* ``CNMAP_MIRROR_V``
+* ``CNMAP_MIRROR`` Same as both ``CNMAP_MIRROR_U`` and ``CNMAP_MIRROR_V``
+
+* ``CNMAP_INVERT_SIGN`` Inverts the sign of the computed normal vector
+
+* ``CNMAP_COMPUTE_OCCLUSION`` Computes a crude occlusion term and stores in the resulting alpha channel
+
+_amplitude_: Scaling factor for normals
+
+_format_: Format of the resulting ScratchImage
+
+# Example
+
+    ScratchImage hmapImage;
+
+    ...
+
+    ScratchImage normalMap;
+    hr = ComputeNormalMap( hmapImage.GetImage(0,0,0), CNMAP_CHANNEL_LUMINANCE | CNMAP_COMPUTE_OCCLUSION, 2.f, DXGI_FORMAT_R8G8B8A8_UNORM, normalMap );
+    if ( FAILED(hr) )
+        ...
+
+# Remarks
+This function does not operate directly on block compressed images. See [[Decompress]] and [[Compress]].
+
+This function cannot operate directly on a planar format image. See [[ConvertToSinglePlane]] for a method for converting planar data to a format that is supported by this routine.
+
+This method computes the normal by using the central difference with a kernel size of 3x3. The central differencing denominator used is 2.0. RGB channels in the destination contain biased (x,y,z) components of the normal.