Markov model is a technique for analyzing events that repeat or extend over a long period of time. It has been widely applied to examine the progressions of disease with well-defined markers such as HIV. Most studies using Markov models suffer from a major limitation of the no-memory assumption. OBJECTIVE: This study uses HIV cases as an example to demonstrate an alternative disease progression model: a history- and duration- dependent transition probability model. METHODS: This general transition probability model captures patients history of disease states (history-dependency) by categorizing HIV patients into three types, stable, progress, and recess, based on whether the patients had the same disease state, a less severe state, or a more severe state at the previous cycle. For patients with a history of stable disease states, duration in that state was also included in the model (duration-dependency). Data from the Multicenter AIDS Cohort Study (MACS) was used to model the disease progression of HIV patients in four U.S. cities since 1984. Five disease states were constructed using ranges of the CD4/CD8 ratio as the marker . They are: > 0.63, 0.42 - 0.63, 0.31 - 0.42, <0.31, and death. The cycle length is 6 months. RESULTS: Compared with estimates obtained from either Markov chain or Markov process models, this general transition probability model yields better prediction of life expectancies and allows more flexibility in modeling the disease progression. CONCLUSIONS: By refining the disease progression process, our transition probability model provides a better way to evaluate the effectiveness as well as cost-effectiveness of treatment alternatives and/or impacts from alternative health policies.