OBJECTIVES: Markov models are commonly used to simulate the natural history of human papillomavirus (HPV) and cervical cancer (CC) to predict health and economic benefits of different prevention strategies. Transition probabilities of moving between health states occasionally cannot be directly estimated from epidemiological/clinical data and sometimes, natural history is not delineated in sufficient detail or data sources may be inaccurate. Nevertheless, the reliability of model outcomes is greatly dependent on accuracy of these inputs. Therefore, a well-calibrated model is essential to ensure credibility of the results. The objective was to assess the impact of most common calibration methods on cost-effectiveness analysis (CEA): manual, Nelder-Mead algorithm and controlled random search (CRS). METHODS: We used a previously published and validated model from Spain. Data targets were age-specific HPV prevalence and CC incidence. Model outcomes included lifetime risk of cancer, quality-adjusted life years (QALYs), and lifetime costs (€). We compared the mean percentage deviation of model-predicted endpoints from available data for the three calibration methods and incremental cost-effectiveness ratios (ICERs) of different CC prevention strategies currently under discussion in Europe. RESULTS: Results showed that with a non-calibrated random matrix, the deviation was 79%. For the manually calibrated matrix, the deviation was 2%, although it required 40 days of analyst work. Regarding automatically calibrated matrices, the deviation was about 7% and 5% with computation times of 25 hours and 100 hours for Nelder-Mead and CRS respectively. Although the most cost-effective strategy remained invariable based in a CEA threshold of 20,000€/QALY, the magnitude of ICERs changed substantially (7,655€/QALY-14,745€/QALY). CONCLUSIONS: Important differences in both goodness of fit and CEA are found depending on the calibration approach. As was expected, the non-calibrated matrices produced HPV prevalence and CC incidence curves very far away from the target values and the largest differences on the cost-effectiveness results.