OBJECTIVES: To examine new methods of incorporating model calibration into sensitivity analyses and the effect of calibration choices on the robustness of model results and uncertainty of model parameters.  METHODS: Seventy-nine model transition probabilities used to describe the natural history of cervical cancer (CC) were calibrated to thirty published target epidemiologic data using Nelder-Mead optimization.  Because factors such as choice of objective function (goodness-of-fit measure) and initial simplex (starting point) may lead to different optimized solutions, we tested the effect of these choices by performing fifteen calibrations using five different objective function weighting schemes and three different initial simplexes.  The objective function was weighted mean percentage deviation calculated by dividing the absolute value of the difference between model estimate and target value by the target value.  Deterministic sensitivity analyses (DSA) were performed by inserting each of the fifteen calibrated parameter sets into the model and assessing the results.   In probabilistic sensitivity analyses (PSA), we assigned an equal probability of selection to each calibrated parameter set and bootstrapped (sampled with replacement) the calibrated parameter sets, combining them with a conventional second-order Monte Carlo simulation for other model parameters.  Model results were assessed using a cost per quality-adjusted life-year (QALY) incremental cost-effectiveness ratio (ICER). RESULTS: The coefficient of variation of individual transition probabilities across the fifteen calibrated parameter sets ranged from 0.06%-162%.  In DSA, the ICER range produced by the calibrations was $6,700-$27,100 per QALY.  When bootstrapped calibrations were included in the PSA, the ICER 95% credible interval was [$6,400,$28,000] compared with [$1,100,$9,400] when using only the best-fitting calibration. CONCLUSIONS: Different calibration methods can lead to different optimized parameter sets, producing different model results.  Model calibration should therefore be incorporated into sensitivity analyses to fully assess the effects of calibration methods on model results.