diff --git a/src/tablib/formats/_csv.py b/src/tablib/formats/_csv.py
index e57e1d9e..46299d12 100644
--- a/src/tablib/formats/_csv.py
+++ b/src/tablib/formats/_csv.py
@@ -18,7 +18,8 @@ def export_stream_set(cls, dataset, **kwargs):
         """Returns CSV representation of Dataset as file-like."""
         stream = StringIO()
 
-        kwargs.setdefault('delimiter', cls.DEFAULT_DELIMITER)
+        if 'dialect' not in kwargs:
+            kwargs.setdefault('delimiter', cls.DEFAULT_DELIMITER)
 
         _csv = csv.writer(stream, **kwargs)
 
@@ -40,7 +41,8 @@ def import_set(cls, dset, in_stream, headers=True, skip_lines=0, **kwargs):
 
         dset.wipe()
 
-        kwargs.setdefault('delimiter', cls.DEFAULT_DELIMITER)
+        if 'dialect' not in kwargs:
+            kwargs.setdefault('delimiter', cls.DEFAULT_DELIMITER)
 
         rows = csv.reader(in_stream, **kwargs)
         for i, row in enumerate(rows):
diff --git a/tests/test_tablib.py b/tests/test_tablib.py
old mode 100755
new mode 100644
index ca8ac05e..e1ae306d
--- a/tests/test_tablib.py
+++ b/tests/test_tablib.py
@@ -1172,6 +1172,38 @@ def test_csv_formatter_support_kwargs(self):
         d2 = tablib.import_set(_csv, format="csv", **kwargs)
         self.assertEqual(3, len(d2.headers))
 
+    def test_csv_import_custom_dialect_delimiter_not_overridden(self):
+        """A custom dialect's delimiter must not be overridden by the
+        CSV formatter's own default delimiter (issue #622).
+
+        ``import_set``/``export_stream_set`` used to unconditionally set
+        ``delimiter=','`` in kwargs before calling ``csv.reader``/
+        ``csv.writer``. Since Python's csv module lets explicit fmtparams
+        override attributes of a passed ``dialect``, this silently
+        overrode any delimiter configured on a custom dialect back to
+        the default comma, even though no delimiter was requested by the
+        caller.
+        """
+        import csv as csv_module
+
+        class ColonDialect(csv_module.excel):
+            delimiter = ":"
+
+        content = "first_name:last_name\nJohn:Adams\nGeorge:Washington\n"
+        d = tablib.import_set(
+            StringIO(content), format="csv", dialect=ColonDialect
+        )
+        self.assertEqual(["first_name", "last_name"], d.headers)
+        self.assertEqual(("John", "Adams"), d[0])
+        self.assertEqual(("George", "Washington"), d[1])
+
+        data.append(self.john)
+        data.append(self.george)
+        data.headers = self.headers
+        exported = data.export("csv", dialect=ColonDialect)
+        self.assertIn(":", exported)
+        self.assertNotIn(",", exported.split("\n")[0])
+
 
 class TSVTests(BaseTestCase):
     def test_tsv_import_set(self):