diff -r 07a6d4c30c4e Lib/test/test_float.py
--- a/Lib/test/test_float.py	Mon Dec 09 01:59:07 2013 +0100
+++ b/Lib/test/test_float.py	Mon Dec 09 14:50:44 2013 +0800
@@ -769,6 +769,16 @@
             test(sfmt, NAN, ' nan')
             test(sfmt, -NAN, ' nan')
 
+    def test_round_with_zero_ndigits_returns_integer(self):
+        x = round(1.23, 0)
+        self.assertEqual(x, 1)
+        # It should return 1 not 1.0
+        self.assertIsInstance(x, int)
+        x = round(4.83, 0)
+        self.assertEqual(x, 5)
+        # It should return 5 not 5.0
+        self.assertIsInstance(x, int)
+
 
 # Beginning with Python 2.6 float has cross platform compatible
 # ways to create and represent inf and nan
diff -r 07a6d4c30c4e Objects/floatobject.c
--- a/Objects/floatobject.c	Mon Dec 09 01:59:07 2013 +0100
+++ b/Objects/floatobject.c	Mon Dec 09 14:50:44 2013 +0800
@@ -979,7 +979,12 @@
     x = PyFloat_AsDouble(v);
     if (!PyArg_ParseTuple(args, "|O", &o_ndigits))
         return NULL;
-    if (o_ndigits == NULL) {
+
+    if (o_ndigits)
+        /* interpret second argument as a Py_ssize_t; clips on overflow */
+        ndigits = PyNumber_AsSsize_t(o_ndigits, NULL);
+
+    if (o_ndigits == NULL || ndigits == 0) {
         /* single-argument round: round to nearest integer */
         rounded = round(x);
         if (fabs(x-rounded) == 0.5)
@@ -988,8 +993,6 @@
         return PyLong_FromDouble(rounded);
     }
 
-    /* interpret second argument as a Py_ssize_t; clips on overflow */
-    ndigits = PyNumber_AsSsize_t(o_ndigits, NULL);
     if (ndigits == -1 && PyErr_Occurred())
         return NULL;