Substantial correlational evidence supports a causal (mediational) interpretation of alcohol expectancy operation, but definitive support requires a true experimental test. Thus, moderately to heavily drinking male college students were randomly assigned to 1 of 3 conditions in a pre-post design: Expectancy challenge (designed to manipulate expectancy levels), "traditional" information, and assessment-only control. Expectancy challenge produced significant drinking decreases, compared with the other 2 groups. Decreases in measured expectancies paralleled drinking decreases in the challenge condition. Significant increases in alcohol knowledge in the traditional program were not associated with decreased drinking. These experimental findings support a causal (mediational) interpretation of expectancy operation. The implications for a cognitive (memory) model of expectancies and for prevention and intervention programs for problem drinking and alcoholism are discussed.