OBJECTIVES: To evaluate the impact, on Brazilian Public Health Care System (SUS) budget, of reimbursing for dasatinib for newly diagnosed CML patients who become imatinib-resistant.  The budget impact analysis is conducted for three consecutive years. METHODS: Due to the rapid evolution of the disease, a monthly-cycle Markov model incorporating clinical and epidemiological data was developed to determine the target population throughout the analysis time. The base case analysis for imatinib-resistant CML patients compared the costs of imatinib (600-800mg/day), versus dasatinib (100-140mg/day). Disease progression depended on the best treatment response rates taken from START clinical trials. Pharmaceutical costs were obtained according to the official price and standard government discounting procedures, but alternative costing scenarios were also evaluated. Data sources for the epidemiological and treatment regimen distributions input variables include reports from the Brazilian National Cancer Institute (INCA) and published literature. Probability sensitivity analysis (PSA) was conducted to account for uncertainties of the model. The three-year time period was considered adequate as approval of nilotinib in the near future will impact the market share structure for imatinib-resistant CML patients for later years. RESULTS: Annual incidence of CML was estimated as 1431. Within one year’s time, 481 CML patients are expected to become imatinib-resistant. In the base case, the net budget impact was a savings of about €1,900,000 in 2009 to savings around €3,300,000 in 2011, with a total savings of approximately €8,000,000/three years time frame. The net impact on the alternative costing scenarios remained negative. The acceptability curve generated from the PSA showed a 100% probability of savings on the base case. CONCLUSIONS: The inclusion of dasatinib as 2nd line therapy for newly diagnosed CML patients who have become imatinib-resistant in Brazil would result in increasingly and significant savings, even after accounting for uncertainties of the model.