This study examines the impact of subject bias in responding to faces formats, Likert formats, and other types of commonly used response formats. Students (N = 292) in a large urban university were instructed to respond randomly to several pages of contentless items. This procedure makes it possible to determine response format bias without contamination from item content. Results indicate that faces formats are subject to considerably more positive bias than Likert and numerical formats and, consequently, that faces formats should not be used in empirical research.