Purpose: Most epidemiological studies of major depression report period prevalence estimates, which are of limited utility in characterizing incidence and understanding the dynamics of this condition. The objective of this study was to integrate data from three recent Canadian studies using a Markov model into a more comprehensive description of major depression epidemiology. Method: Data from national surveys conducted by the Canadian national statistical agency (Statistics Canada) were used. These data were analyzed using a Markov model designed to elucidate the dynamics of depression (previous version: Medical Decision Making 2004 Jul-Aug;24(4):351-8). The model adopts the format of an incidence-prevalence model, modeling the “prevalence pool” of major depression as a function of inflow to the pool (incidence) and outflow through recovery. Changes in health state (depressed and not-depressed) were evaluated over a series of one week stages. To address declining probabilities of recovery with mounting episode length, a Markov tunnel was used to represent the process of recovery. The basic model was expanded to include three recurrence categories. Monte Carlo simulations were used to calibrate model parameters to the observed data. Tracker variables were defined so that the model output could be tracked using definitions comparable to those used in the database. Tracker variables were also programmed to represent the accumulated number of episodes over the entire observation period. Results: The population fell into three categories. A low recurrence group (88%) had a weekly episode incidence (transition probability) of 0.00028. Five percent of the population occupied a moderate recurrence category in which the weekly episode incidence was 0.0010. Seven percent of the population was in a high recurrence category with a weekly episode incidence of 0.00575. The transition probabilities reflecting recovery were high in the initial weeks of the episodes, and declined by an approximately fixed proportion with each passing week. Conclusions: Markov models can provide a mechanism for integration and interpretation of psychiatric epidemiological data. Ultimately, such models may be useful for surveillance and economic analyses of this condition. The methods employed here may also be useful for capturing the dynamics and natural history of other medical conditions.