Understanding the long-term course of problematic drinking is a fundamental concern for health services research in the alcohol field. The stability of, or change in, the course of drinking--especially heavy drinking--has both theoretical and applied relevance to alcohol research. We explore the application of latent class growth modeling to 5 years of survey data collected from dependent and problem drinkers--some not in treatment at baseline--in an attempt to uncover prototypical longitudinal drinking patterns. Results indicated that five profiles of drinkers can be used to represent their longitudinal course of alcohol consumption: early quitters (N = 88), light/non-drinkers (N = 76), gradual improvers (N = 129), moderate drinkers (N = 229), and heavy drinkers (N = 572). Significant baseline factors included ASI drug severity, dependence symptoms, and marital status. Attendance at AA meetings, the size of one's heavy drinking and drug using social network, past treatment, receiving suggestions about one's drinking, and contacts with the medical system were significant influences. The size of heavy drinking and drug using social networks was noticeably larger for the heavy drinkers. Findings also support the usefulness of a semi-parametric latent group-based approach as a tool for analyzing alcohol-related behaviors.