The harsh environment of commercial steel continuous casting processes makes taking measurements difficult, expensive and limited with regard to the information gained. Computational models potentially offer deeper knowledge, but only if they can accurately predict the plant behavior. This requires including and solving the equations which govern all of the important physical phenomena. To achieve reasonable speed while retaining accuracy, computational models must be simplified and calibrated to match plant experiments, using parameters which remain constant over the range of processing conditions of interest. Only after verification, calibration and validation can a model be used reliably as a predictive tool to investigative complex processes such as continuous casting.