Because P300 is typically measured from an average of single trials, variations among individual trials may account for P300 amplitude reduction so often seen in patients with schizophrenia. We tested three hypotheses regarding single-trial contribution to small average P300s in schizophrenics: normal P300s are elicited on some trials and no P300s on others, all trials have consistently small P300s, or P300 latency varies over trials. Nineteen schizophrenics and 35 controls were tested on a two-tone auditory oddball event-related potential (ERP) paradigm. ERPs recorded from the parietal electrode (Pz) were subjected to a P300-screening procedure in which a 2 Hz half-sine wave template was moved across the electroencephalogram (EEG) to find the point of best fit. If, for the point of best fit, the EEG:Template covariance was greater in the signal epoch (280-600 msec) than in the noise epoch (610-930 msec), and if the EEG:Template correlation was statistically significant, the trial passed the P300-screen and was deemed to have a P300. Three types of average ERPs were constructed: Traditional Average from all good (artifact-free, correct response) trials, P300-Screen Average from all good trials that also passed the P300-screen, and Latency Adjusted Average by aligning the P300-screen trials at the latency of maximum covariance. Traditional average ERPs were significantly smaller in schizophrenics than in controls. The results of the P300-screen confirmed all three hypotheses: schizophrenics had fewer trials passing the P300-screen, smaller P300s on each trial, and P300s that were more variable in latency across trials.(ABSTRACT TRUNCATED AT 250 WORDS)