According to the docstring, plt.contour(X, Y, Z, N) should "contour N automatically-chosen levels".

The actual results are somewhat surprising:

import numpy as np
import matplotlib.pyplot as plt

np.random.seed(0)
xx, yy = np.mgrid[-3:3:, -3:3]

for _ in xrange(10):
    zz = np.random.randn(6, 6)
    cset = plt.contour(xx, yy,  zz, 9)
    print(len(cset.levels)),

prints 9 9 8 8 9 9 8 8 9 9.

Additionally, the docstring for plt.contourf has the same prose, but adds some more confusing behavior:

for _ in xrange(10):
    zz = np.random.randn(6, 6)
    cset = plt.contourf(xx, yy,  zz, 9)
    print(len(cset.levels)),

prints 11 10 11 10 10 11 10 11 11 11 (note that the N is the same as for the contour call above).

Am I missing something, or is this is a bug?