diff --git a/beets/library.py b/beets/library.py
index dcc7b3ba0..ba05b69c7 100644
--- a/beets/library.py
+++ b/beets/library.py
@@ -1126,6 +1126,13 @@ class Library(BaseLibrary):
         # Encode for the filesystem, dropping unencodable characters.
         if isinstance(subpath, unicode) and not fragment:
             encoding = sys.getfilesystemencoding() or sys.getdefaultencoding()
+            if encoding == 'mbcs':
+                # On Windows, a broken encoding known to Python as
+                # "MBCS" is used for the filesystem. However, we only
+                # use the Unicode API for Windows paths, so the encoding
+                # is actually immaterial so we can avoid dealing with
+                # this nastiness. We arbitrarily choose UTF-8.
+                encoding = 'utf8'
             subpath = subpath.encode(encoding, 'replace')
 
         # Preserve extension.
diff --git a/beets/util/__init__.py b/beets/util/__init__.py
index d93f9f8a6..ecd66a58f 100644
--- a/beets/util/__init__.py
+++ b/beets/util/__init__.py
@@ -284,12 +284,16 @@ def syspath(path, pathmod=None):
         return path
 
     if not isinstance(path, unicode):
-        # Try to decode with default encodings, but fall back to UTF8.
-        encoding = sys.getfilesystemencoding() or sys.getdefaultencoding()
+        # Beets currently represents Windows paths internally with UTF-8
+        # arbitrarily. But earlier versions used MBCS because it is
+        # reported as the FS encoding by Windows. Try both.
         try:
-            path = path.decode(encoding, 'replace')
+            path = path.decode('utf8')
         except UnicodeError:
-            path = path.decode('utf8', 'replace')
+            # The encoding should always be MBCS, Windows' broken
+            # Unicode representation.
+            encoding = sys.getfilesystemencoding() or sys.getdefaultencoding()
+            path = path.decode(encoding, 'replace')
 
     # Add the magic prefix if it isn't already there
     if not path.startswith(u'\\\\?\\'):
diff --git a/docs/changelog.rst b/docs/changelog.rst
index 1eabcf315..42bcebe60 100644
--- a/docs/changelog.rst
+++ b/docs/changelog.rst
@@ -86,6 +86,11 @@ art for your music, enable this plugin after upgrading to beets 1.0b15.
   already at its destination.
 * Fix Unicode values in the ``replace`` config option (thanks to Jakob Borg).
 * Use a nicer error message when input is requested but stdin is closed.
+* Fix errors on Windows for certain Unicode characters that can't be represented
+  in the MBCS encoding. This required a change to the way that paths are
+  represented in the database on Windows; if you find that beets' paths are out
+  of sync with your filesystem with this release, delete and recreate your
+  database with ``beet import -AWC /path/to/music``.
 
 .. _artist credits: http://wiki.musicbrainz.org/Artist_Credit
 
diff --git a/test/test_db.py b/test/test_db.py
index beafaa6d1..49a1dbef5 100644
--- a/test/test_db.py
+++ b/test/test_db.py
@@ -21,6 +21,7 @@ import posixpath
 import shutil
 import re
 import unicodedata
+import sys
 
 import _common
 from _common import unittest
@@ -428,6 +429,19 @@ class DestinationTest(unittest.TestCase):
         dest = self.lib.destination(self.i, platform='linux2', fragment=True)
         self.assertEqual(dest, unicodedata.normalize('NFC', instr))
 
+    def test_non_mbcs_characters_on_windows(self):
+        oldfunc = sys.getfilesystemencoding
+        sys.getfilesystemencoding = lambda: 'mbcs'
+        try:
+            self.i.title = u'h\u0259d'
+            self.lib.path_formats = [('default', '$title')]
+            p = self.lib.destination(self.i)
+            self.assertFalse('?' in p)
+            # We use UTF-8 to encode Windows paths now.
+            self.assertTrue(u'h\u0259d'.encode('utf8') in p)
+        finally:
+            sys.getfilesystemencoding = oldfunc
+
 class PathFormattingMixin(object):
     """Utilities for testing path formatting."""
     def _setf(self, fmt):