Ross Fadely and I co-wrote a document today laying out the basic hierarchical framework for performing star–galaxy classification in multi-band imaging. It involves a large set of stellar models and a large set of galaxy models. Then a hierarchical approach puts properly normalized and properly informative prior probabilities on all the models given the total data set. Each object then is classified using the prior information inferred from everything, and prior information and classifications are inferred simultaneously. Will it work? I have faith in the Church of Bayes!