College students and police investigators watched or listened to 10 prison inmates confessing to crimes. Half the confessions were true accounts; half were false--concocted for the study. Consistent with much recent research, students were generally more accurate than police, and accuracy rates were higher among those presented with audiotaped than videotaped confessions. In addition, investigators were significantly more confident in their judgments and also prone to judge confessors guilty. To determine if police accuracy would increase if this guilty response bias were neutralized, participants in a second experiment were specifically informed that half the confessions were true and half were false. This manipulation eliminated the investigator response bias, but it did not increase accuracy or lower confidence. These findings are discussed for what they imply about the post-interrogation risks to innocent suspects who confess.