diff --git a/pylearn2/datasets/csv_dataset.py b/pylearn2/datasets/csv_dataset.py
index d7b1845f11..b2cbc48d15 100644
--- a/pylearn2/datasets/csv_dataset.py
+++ b/pylearn2/datasets/csv_dataset.py
@@ -125,6 +125,7 @@ def __init__(self,
         if self.task == 'regression':
             super(CSVDataset, self).__init__(X=X, y=y, **kwargs)
         else:
+            y = y.astype('int32')
             super(CSVDataset, self).__init__(X=X, y=y,
                                              y_labels=np.max(y) + 1, **kwargs)
 
diff --git a/pylearn2/datasets/tests/test_csv_dataset.py b/pylearn2/datasets/tests/test_csv_dataset.py
index 30bc204ef1..3a8d9c519d 100644
--- a/pylearn2/datasets/tests/test_csv_dataset.py
+++ b/pylearn2/datasets/tests/test_csv_dataset.py
@@ -9,7 +9,8 @@ def test_loading_classification():
                              'datasets', 'tests', 'test.csv')
     d = CSVDataset(path=test_path, expect_headers=False)
     assert(np.array_equal(d.X, np.array([[1., 2., 3.], [4., 5., 6.]])))
-    assert(np.array_equal(d.y, np.array([[0.], [1.]])))
+    assert(np.array_equal(d.y, np.array([[0], [1]])))
+    assert (d.y.dtype == 'int32')
 
 
 def test_loading_regression():