OBJECTIVE: The extent to which problem drinkers transition into and out of problem drinking was examined using Markov modeling. METHOD: Study participants (N = 1,350) were randomly sampled from one county’s general population and from consecutive admissions to public and private alcohol treatment programs in the same county, and they were assessed at 1-, 3-, 5-, and 7-year follow-ups. At baseline, all met the criteria for problem drinking. Individuals were classified as ‘problem drinkers’ if they reported at least two of three criteria (heavy episodic drinking, social consequences, dependence symptoms according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition) during the prior 12 months. RESULTS: Although all possible patterns were observed, a latent Markov model with heterogeneous transitions and five patterns fit the data. The sampling frame and baseline alcohol severity related to pattern. CONCLUSIONS: The data indicate that, although they do change over time, problem drinkers on the whole are more likely either to remain problem drinkers or to cease to be problem drinkers than they are to move into and out of problem-drinking status. Once they transition out of problem drinking, they are more likely to remain nonproblem drinkers.