I show that Eric Winsberg’s principles of model-building given in Science in the Age of Computer Simulation are insufficient to argue for the external validation of simulation data in cases in which simulation results conflict, and that laboratory experiments have an advantage over simulations because conflicting experimental results can be decided between on the basis of reproducibility. I also argue that robustness of predictions serves the same function for simulations as repeatability does for laboratory experiments in either adjudicating between conflicting results or allowing us to say that we do not have sufficient justification to validate the results. Finally, I argue for an interpretation of the argument from robustness that appeals to the convergence of many well-built and diverse models rather than the more common interpretation which appeals to the probability that one of a set of models is likely to be true.