A method is developed and applied to select optimal models for loads encountered during the atmospheric re-entry of a spacecraft. In general, information on the re-entry environment is limited, meaning that two or more models for this environment may be consistent with the available information. This defines a collection of candidate models; each model in the collection is consistent with the available information. Methods from decision theory are applied to select the optimal member from the collection. A performance criterion, based on postulated utility functions, is used in the selection process. Herein, we model the re-entry environment as a stochastic process in both space and time. Information on the probability law of the process is limited. The candidate models form a class of non-Gaussian, stationary, stochastic processes. It is shown that the response of a critical internal component is sensitive to assumptions on the unknown properties of the input; the response of the component is therefore used as the performance criterion to select an optimal model for the re-entry environment.