Economic models for highly active antiretroviral (HAART) drug regimens used in HIV infection differ from other models in that they must accommodate the results of short trials, but yield relatively long-reaching predictions of health outcomes, resource use, and costs, in an environment where valid epidemiologic data are scarce. Standard Markov models, were effective for short survival times. With improved treatment options, survival of HIV patients has been prolonged significantly. A methodological consequence is that Markov models with fixed transition probabilities may no longer be reliable. Furthermore, the number and disparity of outcomes of interest, such as surrogate marker changes, primary health outcomes, treatment side effects, degree of adherence, and viral resistance to subsequent regimens complicate Markov approaches. OBJECTIVE: To compare two possible solutions:1) a system of linked, staged Markov models; and 2) a hierarchical Monte Carlo simulation. METHODS: Application of both methods to existing data. FINDINGS: The data requirement for the staged Markov system depends on whether transitions are derived empirically or from theory. We found that approximately 27 exposure years are required per health state to ensure stable incidence densities, and that 12 model states health states can capture both risk of progression and risk of events observed in about 1500 patients over 3 years, but that stable transition rates require at least 50 exposure years per health state, and vary by antiretroviral experience. The hierarchical Monte Carlo model has more modest requirements for data density, but depends on long streams of uninterrupted observations from unique patients. The strengths and weaknesses of these two approaches will be illustrated with data from the ritonavir and ABT378 trials.