Points can be rasterized by setting their zorder attribute to a value less than the rasterization_zorder value of the current axes. However, I cannot get this to work if I set markeredgecolor to 'none'. Here is an example:

import matplotlib.pyplot as plt
ax = plt.subplot(1, 1, 1)
ax.set_rasterization_zorder(0)

# red point, no border
ax.plot(0, 0, linestyle='none', marker='o', markersize=100,
        markerfacecolor='r', markeredgecolor='none', zorder=1)

# rasterized green point, black border
ax.plot(1, 0, linestyle='none', marker='o', markersize=100,
        markerfacecolor='g', markeredgecolor='k', zorder=-1)

# rasterized blue point, no border
ax.plot(2, 0, linestyle='none', marker='o', markersize=100,
        markerfacecolor='b', markeredgecolor='none', zorder=-1)

ax.axis([-0.5, 2.5, -1, 1])
plt.savefig('test.pdf')

Everything works as expected in an interactive window, but the blue point is invisible in the pdf. The only difference between the red point and the blue point is zorder, and the only difference between the green point and the blue point is markeredgecolor. The invisibility of the blue point in the pdf therefore seems to be caused by a combination of rasterization and not having an edge.

I discovered this while running some old plotting scripts I wrote when I was using matplotlib 1.3 installed via pip for homebrew python on Mac OS 10.6.8. I'm now using anaconda python for Mac OS 10.9, which comes with matplotlib 1.4.1. I feel like there must be some connection with either the updated matplotlib or the anaconda python distribution (after all, my old scripts used to produce the correct output), but I cannot test this without reverting my entire python installation.

Does anyone know what could be going on here?