This paper examines the act of disagreement as realized in university settings and the choice of linguistic markers used to soften or strengthen disagreement. It is based on a corpus of natural data containing over 450 turns in which disagreement was expressed in university courses and academic talks. Brown and Levinson's (1987) factors of power and rating can be used to understand some of the ways in which disagreement was expressed in the natural data, but these factors cannot be treated as formulaic variables, nor are they sufficient to account for all means for expressing disagreement. Relative power of interlocutors was relevant in that the professors observed here used more markers of positive politeness when disagreeing with students than did peers disagreeing with each other or students disagreeing with professors. Severity of disagreement (Brown and Levinson's rating of imposition) could have two opposite results: in one, politeness was increased to lessen face threat to the addressee; in the other, face threat to the speaker outweighed considerations of the addressee's face, leading to aggravated disagreement. Face maintenance concerns, however, do not account for how disagreement was used to serve pedagogical aims, specifically when disagreement was used as part of the elicitation sequence.