Forty subjects participated in a four-choice reaction time experiment in which they received alcohol (1 ml/kg body weight) in one session and no alcohol in another on consecutive days (the order being counterbalanced). Fifty practice trials and then 2,000 experimental trials were given in each session. Subjects were slower, more variable, and less accurate overall with alcohol than without. They were also slower and more variable in the first than in the second session. By combining the data across subjects to produce speed-error tradeoff functions, it appears that practice has little effect whereas alcohol has a substantial effect; responses made within 600 msec of stimulus presentation are more likely to be erroneous with alcohol than without. In addition, the number of consecutive errors was increased by alcohol. The results are interpreted within a model in which it is suggested that the primary effect of alcohol upon performance in speeded tasks is to decrease the rate of accumulation of evidence. © 1987 Psychonomic Society, Inc.