Using a 3-wave longitudinal design, adolescents were studied over a 2-year period during which many first began to drink. Covariance structure modeling showed that teens' expectancy for social facilitation from alcohol and their drinking experience influenced each other in a reciprocal, positive feedback fashion: the greater the expectancy endorsement, the higher subsequent drinking levels, and the higher the drinking levels, the greater the subsequent expectancy endorsement. This model fit the data quite well; comparison models, in which expectancy (or drinking) had no independent influence on future drinking (or expectancy), showed significantly poorer fit than the present model. Initial nondrinkers' social expectancy predicted individual differences in the rate of drinking increase over the 2 years. Results bolster the hypothesis that expectancy actively influences drinking and point to the importance of expectancy-based intervention efforts.