If multiple etiologies of substance use are truly at work in the population, then further strides in the accurate prediction of smoking and the use of other substances will likely be built on diverse pattern-centered approaches that explore the presence of multiple population subgroups across various substance use stages. The present study aimed to identify population subgroups defined by individual risk factors or risk factor constellations that prospectively predict specific smoking stages. Using data from the National Longitudinal Study of Adolescent Health (Add Health), analyses were conducted on the sample that took part in the baseline and 1 year follow-up assessment between 1994 and 1996. Classification and regression tree procedures were used to investigate the structure of individual risk factors, or constellations of risk, that define population subgroups with high rates of both experimental and established smoking. For each level of smoking, a relatively simple model including two subgroups predicted over half of the smoking cases. Findings also indicated that the two group models identified higher rates of regular smokers compared to experimental smokers. Deviant behaviors and alcohol use without permission independently predicted movement to experimentation at follow-up. Progression to regular smoking from both a nonsmoking and experimental smoking status at baseline were each predicted by smoking friends. Additionally, baseline levels of experimental use predicted movement from experimental to regular smoking, while a relatively low grade point average predicted rapid progression from baseline nonuse to regular use at follow-up. By identifying first approximations of patterns, these analyses may lead to clues regarding the major multiple mechanisms at work for the progression of smoking among adolescents.