This dissertation discusses results obtained through formulation and estimation of a dynamic stochastic model that captures individual smoking decision making, health expectations, and longevity over the life cycle. The standard rational addiction model is augmented with a Bayesian learning process about the health marker transition technology to evaluate the importance of personalized health information in the decision to smoke cigarettes. Additionally, the model is well positioned to assess how smoking and smoking cessation impact morbidity and mortality outcomes while taking into consideration the potential for dynamic selection of smoking behaviors. This research also provides a novel approach to the empirical construction of the theoretically common ""smoking stock"" that facilitates the estimation of investment and depreciation parameters. The structural parameters are estimated using rich longitudinal health and smoking data from the Framingham Heart Study: Offspring Cohort. Results suggest that there exists heterogeneity across individuals in the pathways by which smoking effects health. Furthermore, upon smoking, the estimated parameters suggest a positive reinforcement effect and a negative withdrawal effect, both of which encourage future smoking. I find that only in the case of very large change in an individual's health markers will the associated change in beliefs induce individuals to quit smoking. Generally, personalized health marker information is not found to influence smoking behavior relative to chronic health shocks themselves. The dissertation also presents evidence of health selection in smoking behavior that, when not modeled, may cause an overstatement of the direct effect of smoking on morbidity and mortality.