Varanid lizards are difficult to sex in the field because commonly used techniques are not completely reliable and definitive techniques are not logistically or economically feasible for many field-based applications. Previous work has shown that variation in morphometric variables can be used to determine sex in some species of varanid. Here we build on these previous exploratory analyses by developing a set of a priori models (containing morphometric variables) to predict the sex of six species of Australian varanid, and then examining their relative support under the information-theoretic framework. We then use cross-validation procedures to determine the reliability of the best-supported models' predictive ability. Our analysis suggests that a large sample size is required for building models to predict sex in many species. The most important sexually diagnostic features for many species were a number of head variables and (to a lesser extent) scaling of limb proportions. This analysis provides some useful statistical tools for the field-sexing of adult and juvenile Varanus gouldii with a known level of reliability and also serves to highlight the danger of extrapolating from potentially spurious results when using exploratory methods or null hypothesis testing.