The authors investigated whether observers infer others' credibility from the consistency of their visible and audible characteristics, and whether such inferences are justified. In Study 1, target persons were videotaped while reading a standard text; in Study 2, target persons were videotaped while lying or telling the truth. From these videotapes, silent films and audiotapes were produced and presented to independent observers who inferred the targets' personality traits from this information. Measures of cross-modal discrepancy were derived from differences between personality descriptions based on a silent film or an audiotape. Lying resulted in cross-modal discrepancies in impressions of Agreeableness, and cross-modal discrepancies in Agreeableness were related to judgments of dishonesty. Deception detection was substantial if the judges were exposed to acoustic information on the targets, and if the targets faked their curriculum vitae. Deception detection was to some extent, but not entirely, mediated by cross-modal discrepancies in impressions of Agreeableness.