Letters of recommendation typically use subjective language that is open to interpretation. In three studies, participants “translated” letters of recommendation for female and male applicants written by sexist, anti-sexist, or “control” professors. Predictions were based on the shifting standards model [Biernat, Manis, & Nelson, 1991], the attributional principle of augmentation [Kelley, 1971, and models of “correction” for contamination [e.g., Wilson & Brekke, 1994]. Participants translated equivalent letters as indicating lesser ability in female than male applicants, particularly when the letter writer was described as sexist, but own impressions of the candidate diverged from those of the sexist writer. Differential standard use and attributional augmentation did not appear to be responsible for these effects. Instead, writer sexism triggered dislike and corrective processes in impression formation. Copyright © 2007 John Wiley & Sons, Ltd.