All models are based on mathematical expressions that depend on parameters (sometimes hidden, as in the case of neural networks), whose optimal value needs to be found to maximize the accuracy of its predictions. Traditionally, the value of a model parameters has been found by minimizing the error between experimental observations and model predictions. This strategy, often requiring ad-hoc regularization, provides point values for the calibrated parameters but gives no information with regards to their uncertainty and might, moreover, correspond to local optima. A more powerful model calibration methodology can be designed following the ideas of Bayesian inference, pioneered by Kennedy and O'Hagan [1], and later adopted by diverse fields such as medicine, energy efficiency, material science, etc. In this talk I will review Bayesian model calibration and apply its ideas to calibrate a material model that depends on more than a dozen parameters and represents its thermo-mechanical response in the inelastic regime [2] subjected to high-strain rate and high-temperature conditions. The focus will not be on the material itself, but rather on the step-by-step process that is required to calibrate a complex model and the valuable insights that can be gained from this approach.