In this paper, we present a Markov model for estimating the cost-effectiveness of combination therapy with lamivudine (LMV) and zidovudine (ZDV) compared with ZDV alone. We also compare the predictions of the Markov model for the impact of combination therapy on trial period costs with the actual impact of combination therapy on selected trial period costs estimated from data collected during the clinical trials. In the Markov model, disease stages were defined by CD4 cell count. Based on clinical trial data for patients with CD4 counts higher than 100 cells/mm3, the model assumed that the CD4 cell count level could be maintained above the level at the initiation of therapy for 6.5 months with monotherapy and for 18 months with combination therapy. After this period, transition rates for natural disease progression were used. Incremental lifetime costs and quality-adjusted life years gained with LMV/ZDV compared with ZDV alone were estimated for cohorts of patients initiating antiretroviral therapy at four different CD4 cell count stages. Cost per life year gained varied from $10,000 to $18,000, and cost per quality-adjusted life year gained varied from $14,000 to $27,000. In both cases, the combination therapy was more cost-effective when started earlier in disease progression. These estimates were not sensitive to changes in key parameter values. In addition, the model was used to estimate the impact of combination therapy on healthcare costs during the trial period; these estimated costs were compared with data on the cost of resource use collected during the clinical trial for hospital stays, unscheduled visits, medications, and outpatient procedures. Both the Markov model estimates and the trial data estimates for the trial period showed cost savings in other medical costs, though these were not large enough to completely offset the increased cost for antiretroviral therapy. The model estimates were more conservative than the estimates based on the trial data.