A new method was developed to compare two models using vague priors, showing they lead to infinite Bayes factors. For linear models, certain improper priors can give a finite Bayes factor, with a subclass making it related to the F-statistic. This was demonstrated in a one-way analysis of variance scenario.