The present work establishes a methodology used to conduct a comparative study of various probability distributions generally used to describe liquid spray atomization data. This methodology involves model discrimination and criticism, as well as random sampling used to obtain objective criteria for choosing among rival models. Examples, representing different atomization mechanisms, are given for the determination of the relative posterior probabilities of candidate models and for their goodness of fit when models are fitted to a common data set. The methodology has been extended to include unreplicated data in addition to replicated data. The end result Is presented in a form suitable for formulating empirical models of spray size distributions, with the intent of incorporating such models into multidimensional, two-phase flow simulations.