Model uncertainty is well-known as an almost inevitable feature of wildlife studies.  Legitimate model-based inference requires at least an acknowledgment of this uncertainty, though what is really desired is an assessment of its consequences. 
Assessing model uncertainty requires a model set and a procedure for evaluating the relative support provided by data for the various models.  The Bayesian paradigm provides a natural framework for multimodel analysis; indeed, I argue that model averaging is inevitably Bayesian.
My purpose in so stating is not to promote silly Bayesian/frequentist bickering, but rather to encourage a close look at what might otherwise remain latent issues explaining the performance of model-weighting schemes.  In particular, I offer an explanation for the widely noted tendency for AIC weights to favor complex models. 
In analysis of a single model, it is widely known that the choice of priors on parameters is of little consequence provided that sample sizes are adequate.  Perhaps less widely known is that the choice of priors on parameters can have serious consequences for multimodel Bayesian inference.  I illustrate potential problems and discuss principles for selecting priors on parameters so as to allow an objective assessment of model support in multimodel Bayesian inference.