This study investigated the validity of procedures for estimating the P3 complex in single trials. In "pseudo-real" simulations of the N1-P2 complex of the occipital visual-evoked potential, Möcks, Köhler, Gasser, and Pham (1988) had reported that their maximum-likelihood method (Pham, Möcks, Köhler, & Gasser, 1987) performed better than Woody's (1967) method. Using pseudo-real simulations of auditory oddball data, we wanted to know whether this finding also held true for the P3 complex. The performance of three methods was studied: peak picking, Woody's method, and Pham et al.'s method (as well as an extension of this latter method). Performance of all methods critically depended on the signal-to-noise ratio. There was some advantage for the more sophisticated methods, particularly when signal-to-noise ratios were realistic. "Good" trials may be selected by all methods, to improve the signal-to-noise ratio, but this selection entails the risk of bias. Further research should investigate whether these conclusions also hold true when the P3 complex consists of more than one component.