BACKGROUND Most formal treatment programs recommend Alcoholics Anonymous (AA) attendance during treatment and as a form of aftercare, but we know very little about treatment seekers' patterns of AA involvement over time and how these relate to abstinence. METHOD This paper applies latent class growth curve modeling to longitudinal data from 349 dependent drinkers recruited when they were entering treatment and were re-interviewed at one or more follow-up interviews one, three and five years later, and who reported having attended AA at least once. RESULTS Four classes of AA "careers" of meeting attendance emerged: The low AA group mainly just attended AA during the 12 months following treatment entry. The medium and high AA groups were characterized by stable attendance at the second and third follow-ups-at about 60 meetings a year for the medium group and over 200 meetings per year for the high group, followed by slight increases for the medium group and slight decreases for the high group by year five. The declining AA group doubled its meeting attendance postbaseline, to almost 200 meetings during the year following treatment entry, but by year five they were only attending about six meetings on average. Decreases in AA meetings did not necessarily signal disengagement from AA; at the five-year follow-up, a third of the low AA group and over half of the declining AA group said they felt like a member of AA. Activities other than meeting attendance, such as having a sponsor, otherwise paralleled the meeting careers, but social networks were similar by year five. Rates of abstinence by year five (for the past 30 days) were 43% for the low AA group, 73% for the medium group, 79% for the high group and 61% for the declining group. Rates of dependence symptoms and social consequences of drinking did not differ between the groups at year five. CONCLUSIONS The prototypical AA careers derived empirically are consistent with anecdotal data about AA meetings: some never connect; some connect but briefly; and others maintain stable (and sometimes quite high) rates of AA attendance. However, contrary to AA lore, many who connect only for a while do well afterwards.