Results of a prospective examination (N = 618) of factors associated with smoking relapse are reported. At 1-year follow-up, a modified version of the Fagerstrom Tolerance Questionnaire (Dependence Index; DI) and a measure of craving entered the logistic model (odds ratio of 2.7 [p less than .001]). At Year 2, only the DI entered the model (odds ratio of 2.2 [p less than .001]). The ability of signal detection analysis (SDA) to produce clinically useful decision rules was also examined. At Year 1, SDA produced 1 subgroup with a 25% nonrelapse rate and another with a 9% nonrelapse rate (odds ratio of 3.4 [p less than .001]). At Year 2, SDA produced 1 subgroup with a nonrelapse rate of 19% and another with a nonrelapse rate of 7% (odds ratio of 3.0 [p less than .001]). The use of signal detection methods may help clinicians to identify those at greater or lesser risk of relapse.