A theory is offered which proposes that alcohol induces eccentricity in the neurophysiological mechanism by which responses are selected for emission by the nervous system. Hence, in a specific stimulus situation, normally high probability responses decrease in frequency, while lower probability responses increase in frequency. From this postulate, the prediction is derived that in a verbal paired-associate learning task administered over two days, alcohol will decrease proactive interference from first-day learning (high probability associate) and result in enhanced acquisition of second day items. This paper reports two experimental tests of the prediction, each of which employs four subject groups tested over two days. Experiment 1 ( n = 24) provides tentative support while ruling out some potential confounds. With these confounds eliminated, Experiment 2 ( n = 40) confirms the prediction in a more direct test. A variety of possible alternative explanations for the counter-intuitive finding of improved cognitive performance under alcohol are evaluated against the theory proposed, as are other recent theories which address the variable behavioral effects of alcohol. Finally, the new data gathered on the effects of imagery instructions on proactive interference are considered.