Radiation imaging systems provide a wealth of information for analyzing quantities of interest (enumeration, strength, dimensions) of an uncharacterized source. This capability is useful, for example, when measuring plutonium powder “held up” in unknown locations at reprocessing facilities. Accurate models of the imager are required to infer the quantity of interest. This paper describes the modeling of a fast-neutron coded aperture imager. The model consists of two components: a forward model (to predict counts from a known source distribution) and a set of covariance models (to predict prediction–experiment residuals). The models are calibrated and cross-validated using 15 measurements of a californium-252 neutron source. The cross-validation results suggest that the models are imperfect but can be used for source-intensity inference.