Decisions on DSM-IV criteria for alcohol dependence were based in part on latent structure analyses of field survey data on alcohol problems. Analogously, to investigate the latent structure of nicotine dependence in an epidemiological sample, we carried out a dichotomous item factor analysis of DSM-III-R symptom data gathered from 394 young adults who reported a history of sustained daily smoking. Smokers and their dependence symptoms were identified by means of the NIMH Diagnostic Interview Schedule version III revised, administered to a random sample of 1007 21–30-year-olds who were members of a health maintenance organization in the Detroit area. Comparing different latent structure models using LISCOMP software with bootstrap re-sampling, followed by multiple logistic regression, vie found that a two-factor model indicating a 'general dependence' and a 'failed cessation' dimension best accounted for the observed data. Current smoking status (persistent vs. past smoking) was associated with the two factors independently. Replication and additional research on construct, discriminant and convergent validity are needed.