# in each case returns array of bool np.isnan(a).any(axis=0) # columns where any value is nan np.isnan(a).all(axis=0) # columns where all values are nan