diff --git a/picard/i18n.py b/picard/i18n.py
index 61c38c306b..73cdcb3176 100644
--- a/picard/i18n.py
+++ b/picard/i18n.py
@@ -245,4 +245,8 @@ def sort_key(string, numeric=False):
     # scripts. Replace numbers in the sort string with their latin equivalent.
     if numeric and (IS_MACOS or IS_WIN):
         string = re.sub(r'\d', _digit_replace, string)
-    return collator.sortKey(string.replace('\0', ''))
+
+    # On macOS numeric sorting of strings entirely consisting of numeric characters fails
+    # and always sorts alphabetically (002 < 1). Always prefix with an alphabeticcharacter
+    # to work around that.
+    return collator.sortKey('a' + string.replace('\0', ''))
diff --git a/test/test_i18n.py b/test/test_i18n.py
index 0d1d405079..41f0b0e15d 100644
--- a/test/test_i18n.py
+++ b/test/test_i18n.py
@@ -82,7 +82,11 @@ def test_sort_key(self):
         i18n.setup_gettext(localedir, 'de')
         self.assertTrue(i18n.sort_key('äb') < i18n.sort_key('ac'))
         self.assertTrue(i18n.sort_key('foo002') < i18n.sort_key('foo1'))
+        self.assertTrue(i18n.sort_key('002 foo') < i18n.sort_key('1 foo'))
         self.assertTrue(i18n.sort_key('foo1', numeric=True) < i18n.sort_key('foo002', numeric=True))
+        self.assertTrue(i18n.sort_key('004', numeric=True) < i18n.sort_key('5', numeric=True))
+        self.assertTrue(i18n.sort_key('0042', numeric=True) < i18n.sort_key('50', numeric=True))
+        self.assertTrue(i18n.sort_key('5', numeric=True) < i18n.sort_key('0042', numeric=True))
 
     def test_sort_key_numbers_different_scripts(self):
         i18n.setup_gettext(localedir, 'en')