BACKGROUND: was to examine the reliability of the algorithm. METHODS: As part of a randomized controlled trial, 3,930 adolescents completed a paper version of the algorithm questions and a differently worded computerized version on the same day: a parallel form reliability assessment. In a separate assessment, another group of 118 adolescents completed 2 identical paper versions of the same questionnaire 2 weeks apart: a test-retest reliability assessment. Kappa (kappa) for agreement for stage and the individual questions were calculated. Logistic regression was used to examine whether demographic characteristics, smoking status, and stage predicted agreement for stage. RESULTS: Kappa (95% confidence intervals) for stage was 0.57 (0.55-0.60) in the first assessment and 0.46 (0.28-0.63) in the second assessment, indicating moderate reliability. The question concerning trying smoking in the next 6 months was moderately reliable, but that concerning trying within the next thirty days was poorly reliable. Acquisition precontemplation was significantly more reliably coded than all other stages. Demographic characteristics did not predict reliability. CONCLUSIONS: The algorithm reliably allocates individuals into acquisition precontemplation, but for all other stages, its reliability is fair.