diff --git a/proselint/checks/mau_a_vs_an.py b/proselint/checks/mau_a_vs_an.py
new file mode 100644
index 000000000..b34c54e11
--- /dev/null
+++ b/proselint/checks/mau_a_vs_an.py
@@ -0,0 +1,70 @@
+# -*- coding: utf-8 -*-
+"""MAU100: a vs. an
+
+---
+layout:     post
+error_code: MAU101
+source:     Garner's Modern American Usage
+source_url: http://amzn.to/15wF76r
+title:      a vs. an
+date:       2014-06-10 12:31:19
+categories: writing
+---
+
+The first line is always wrong.
+
+"""
+import re
+from nltk.corpus import cmudict
+from proselint.memoize import memoize
+
+
+def check(text):
+
+    error_code = "MAU101"
+    msg_a = "'a' should be 'an'."
+    msg_an = "'an' should be 'a'."
+
+    dic = cmudict.dict()
+
+    @memoize
+    def starts_with_vowel_sound(word):
+        """Does the word start with a vowel sound?"""
+
+        # Get the pronunciations of the word.
+        pronunciations = dic.get(word)
+        if pronunciations is None:
+            return None
+
+        # For each pronunciation, see if it starts with a vowel sound.
+        is_vowel = [p[0][-1].isdigit() for p in pronunciations]
+
+        # Return the appropriate value only if all the pronunciations match.
+        if all(is_vowel):
+            return True
+        elif not any(is_vowel):
+            return False
+        else:
+            return None
+
+    errors = []
+    regex = re.compile("(^|\s+)(A|a|An|an)\s(\S*)\W")
+
+    # Find all occurences of the regex in the text.
+    for m in regex.finditer(text):
+        words = [group for group in m.groups()]
+
+        vowel_sound = starts_with_vowel_sound(words[1])
+
+        if vowel_sound is None:
+            continue
+
+        # A apple.
+        if words[0] in ["A", "a"] and vowel_sound:
+            errors.append((m.start(), m.end(), error_code, msg_a))
+
+        # An day.
+        elif words[0] in ["An", "an"] and not vowel_sound:
+            errors.append((m.start(), m.end(), error_code, msg_an))
+
+    return errors