This paper considers simulation models in applied science from a statistical perspective, points out potential statistical pitfalls in calibration, and opportunities for the use of statistical regularities to reduce computational requirements and provide a foundation for systematic testing and evaluation. Sections of the paper deal with the impact of numerical approximation on the statistical properties of estimators, the use of simulation methods to assist statistical inference, and the use of nested multinomial approximations for computational tractability and robustness. † Presenting author * Angelika Eymann died in December 2000, and this paper is dedicated to her memory.