On each trial, one of three stimulus events occurred: Presentation of (1) a 500-cps signal added to a sample of white noise; (2) an 1100-cps signal added to the noise; or (3) no signal added to the noise. In the detection-recognition condition, each listener reported whether a signal occurred (detection) and which of the two signals it was (recognition). The listeners' criteria for a YES response were manipulated through instructions. The functions of primary concern were those relating the proportion of correct recognition responses to the proportion of YES responses on signal-plus-noise trials, p(YES|S). The ability of listeners to detect the signals when they were not required to make the recognition response was measured in the detection condition. The proportion of correct recognition responses on trials on which the detection response was NO, p(C|NO), was above chance at all detection criteria. The data contradict the predictions made by a threshold model adapted to the detection-recognition situation. The proportion of correct recognition responses on all signal trials, whether the detection response was YES or NO, p(C), was unaffected by changes in criterion.