The Restrained Drinking Scale (RDS) was designed to measure an individual's preoccupation with controlling alcohol intake. Cluster analyses indicated that the original RDS could be divided into four subscales. One of the subscales measured frequency and occasions for heavy drinking, and was therefore conceptually and statistically confounded with self-reported consumption. The remaining three subscales of the differentiated RDS outperformed two undifferentiated (with and without the confounding items) versions of the RDS in prediction of drinking behavior. Negative experiences interacted with drinking restraint and contributed in their own right to prediction of drinking. It was concluded that drinking restraint, as measured by the differentiated RDS, encompasses components that relate uniquely to drinking behavior.