As promised, the write-up on prejudice. To simplify things I will stick to the theory that, by all accounts, have much to offer studies on prejudice: the dissociation model developed by Devine.
Devine distinguishes between (1) stereotypes and (2) personal beliefs (attitudes) about group members. She explains that they are different types of stored information. The stereotype is the knowledge of attributes associated with some group, whereas prejudiced personal beliefs are the acceptance and affirmation of the content of a negative cultural stereotype.
This distinction is important because, according to Devine, different cognitive processes govern their activation: automatic processes govern activation of stereotypes, and controlled processes govern the activation of beliefs.
What does it mean that different cognitive processes govern them? Simple: in situations where there is little time, or one has little cognitive capacity, one's heuristic knowledge of a stereotype is more likely to be activated. I.e., even a person low in prejudice may respond with the stereotype, appearing highly prejudiced. Indeed, research found that high and low prejudice subjects responded much the same in such situations. Only when they were given time to convey their responses (think them over, etc.) did they respond according to their personal beliefs. In the case of low prejudice subjects, prejudice was then often in little or no evidence, and egalitarian beliefs frequently apparent.