There's an interesting study on the AP poll, as reported in the WSJ's Econ Blog  Money quote:

The conventional wisdom offered by pundits regarding college football has little empirical support -- and in some instances is exactly the opposite of what "everyone knows" is true," he writes. "Instead of teams suffering more for losing later in the season, a late season loss actually cushions the blow. Rather than being rewarded for defeating strong opponents, AP voters seem not to pay attention to the strength of a defeated opponent. Similarly, margin of victory seems to matter little in how AP voters choose to reward a team's performance.

The question is, if the conventional wisdom is wrong, what do they actually base their votes on?

The pdf version is available here.