OBJECTIVE: To systematically review published economic modeling studies directly comparing specific antiretroviral (ARV) drug treatment strategies. METHODS: A systematic literature search using MEDLINE was conducted of the English-language literature published between 1995 and 2002 to capture modeling studies after introduction of HAART using the following keywords: AIDS, HIV, ARV, modeling, and costs. Manual searches were also conducted of article bibliographies and other relevant databases. Articles were excluded if they were not direct drug comparisons (general HAART vs. no HAART; PI class vs. NNRTI class). Factors including model approaches, use of probabilistic sensitivity analysis, and target audiences and journals were assessed. RESULTS: Of >400 abstracts screened, only 38 articles were identified as economic models of ARV therapy and were reviewed in detail. Ten of these studies (6 US; 2 UK; 2 multi-country) directly compared specific ARV treatments. Of these, 5 assessed cost-effectiveness over timeframes of 20-40 years, of which 2 also evaluated cost-utility. The other 5 presented cost-consequence or cost analysis with timeframes of 1 year in 4 of the studies. Six studies compared zidovudine monotherapy to specific combinations. Four studies used Markov modeling techniques, while 3 used Monte Carlo simulation to test uncertainty. Target audiences were typically the payer, with one of the articles specifically tailored to a state AIDS Drug Assistance Program perspective. Seven of the studies were published in health economic or managed care journals, and 3 published in therapeutic area-specific, clinical journals. CONCLUSIONS: Economic models comparing specific ARV treatments only represent ~25% of published modeling studies in HIV/AIDS. Although antiretroviral (ARV) therapies have been broadly reimbursed in the past, as more treatment options become available, payers may increasingly require economic comparison of specific ARV drug treatment strategies. Until recently, there has been widespread reimbursement of HAART, which may explain the small number of head-to-head economic modeling studies.