The spatial distribution of a cell’s components tells us much about its state and function. CellOrganizer is an open source software package that learns generative models of nuclear shape, cellular shape, and protein location from cell images and can synthesize in silico instances from these models. Our research involved developing methods to evaluate a model’s fit for a given dataset. First, we developed a method for evaluating the reconstruction error of a model’s parameterization for nuclear shape, cellular shape, and protein localization. Second, we developed a method for comparing models of nuclear shape and cellular shape by finding the likelihood of an image set given the generative model. These evaluation methods will provide users of CellOrganizer greater analytic power in their biological research by giving them a toolkit to choose the best fit model.