In this presentation we will begin with an exploration of a generalization of a spatial Poisson mixture. This model is useful for classifying observations to disease classes (for example) that are correlated within spatial regions between class labels, and within class labels between spatial regions. We will also present preliminary results exploring the most appropriate Deviance Information Criterion to use for model selection within this context. While the Deviance Information Criterion (DIC) is a common tool used for model selection within a Bayesian framework, we will focus on a subset of the 8 alternative formulations presented by Celeux and others (2006).