You are being too generous on pseudo-element parsing. There can't be a space before the ":after". See the above test, selector "P.four :after". Other problems on this page are in bug 4833.

To be exact, P.four :after ...should be parsed as meaning the same as: P.four *:after ...i.e., it refers to the ":after" pseudo-element of *every* descendant of any P elements with class=four. It does _not_, however, refer to the :after pseudo-element of the P.four itself, which I guess is what is being done now (I have not checked, though).

Yes. I now think Ian is right. I was looking at 5.10, which says: Pseudo-classes are allowed anywhere in selectors while pseudo-elements may only appear after the subject of the selector. However, 5.3 says (I didn't see before how general this statement is) If the universal selector is not the only component of a simple selector, the "*" may be omitted. This is *not* what you are currently doing. I tested this by adding a "P.four:after" rule after the "P.four :after" rule, and it overrode the first. This applies to other pseudo-elements and pseudo-classes. That is, DIV :first-line == DIV *:first-line :first-letter == *:first-letter (which only influences block-level elements) P :before == P *:before P :link == P *:link A :link == A *:link (useless?) etc.

******** IGNORE EVERYTHING ABOVE THIS LINE. ******** The problem is not parsing. The problem is in matching. You are matching descendant selectors of the form "A *:pseudo-element" as matching the :pseudo- element of A or a descendant of A rather than only descendants of A (and not A itself). This shows up in the following test: http://www.fas.harvard.edu/~dbaron/csstest/pseudos in the selectors div.two :first-line div.six :before As described in the test, the first line of the second test paragraph should not be purple, and the words "Beginning of an Element" should not appear at the beginning of the eighth paragraph. cc:ing troy. I'm not sure who should deal with this.

Fixed. Now "foo:bar" turns into "foo > :bar" and "foo :bar" turns into "foo > *:bar".

Verified fixed.