By selecting the 0th column in particular, as you've noticed, you reduce the dimensionality: >>> m = np.random.normal(0, 1, size=(5, 2)) >>> m[:,0].shape (5,) You have a lot of options to get a 5x1 object back out. You can index usin