OBJECTIVES: Markov models are commonly used to study time dependent disease progression. While most models are cohort based, due to their limitation of dealing with heterogeneities, continuous variables, and dynamic strategies, individual-based microsimulation is being increasingly used. The objective was to compare microsimulation to cohort approach in modeling COPD, while validating the two approaches against findings from TORCH trial. METHODS: We developed both models to study COPD progression in a cohort defined by the characteristics of TORCH patients. The microsimulation randomly generated a large number of patients and tracked each patient’s the lung function (FEV1), exacerbations, and mortality, based on individual’s characteristics and disease history. The cohort model included four COPD stages and death; it modeled exacerbations as events, assuming no impact on transitions or future exacerbations. Both models were populated by published data and results were compared against TORCH findings. RESULTS: The mean decline in FEV1 over 3-year was 126 ml in microsimulation, 49 ml in cohort model, compared to 117 ml in TORCH. The annual rates of moderate and severe exacerbations were 0.94 and 0.18 in microsimulation, 1.12 and 0.18 in cohort model, compared to 1.13 and 0.19 in TORCH. The 3-year mortality was 17.4% in microsimulation, 12.2% in cohort model, and 15.2% in TORCH. Microsimulation required simulating at least 3500 patients to obtain stable estimates, which took 4 minutes to run each scenario. It would take 300 hours to run 5000 scenarios for sensitivity analysis, while the cohort model took less than 1 minute. CONCLUSIONS: In COPD, patient heterogeneity and disease history can be conveniently captured in microsimulation, while parameter uncertainties are easily assessed using cohort approach. The cohort approach is simple to develop, but its inherent Markovian property cannot fully represent COPD pathology. Microsimulation is flexible in mimicking COPD progression, but it is computational expensive.