diff --git a/doc/extra-intersphinx.txt b/doc/extra-intersphinx.txt
index 67f1aef327adf7f7861353747ad0340e5def6c41..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 100644
--- a/doc/extra-intersphinx.txt
+++ b/doc/extra-intersphinx.txt
@@ -1 +0,0 @@
-bob.io.matlab
diff --git a/doc/guide.rst b/doc/guide.rst
index 10799a90a4b34c9828fa3ecdbca0ec305c933fd4..8bc3d8fa6c1855ab0540b664fd4afe3443dddd22 100644
--- a/doc/guide.rst
+++ b/doc/guide.rst
@@ -76,8 +76,7 @@ Once a precise input is given in double-precision gray-scale, both bindings (our
 
 This means that you should expect precision problems if you feed in videos or
 lossy input formats such as JPEG images.  If you input HDF5 files (e.g. using
-:py:mod:`bob.io.base`), Matlab ``.mat`` files (e.g. using
-:py:mod:`bob.io.matlab`) or any other data in formats which are **not** subject
+:py:mod:`bob.io.base`) or any other data in formats which are **not** subject
 to lossy compression/decompression, this data is pre-grayscaled **and** stored
 in double-precision floating point numbers, the output is consistently the
 same, no matter which environment you use.