Discriminant function analysis (DFA) based on morphological measurements is a quick, inexpensive, and efficient method for sex determination in field studies on cryptically monomorphic bird species. However, behind the apparent standardization and relative simplicity of DFA lie subtle differences and pitfalls that have been neglected in some studies. Most of these concerns directly affect assessment of the discriminant performance, a parameter of crucial importance in practice because it provides a measure of the quality of an equation that may be used in later field studies. Using results from 141 published studies and simulations based on a large data set collected on adult Zenaida Doves (Zenaida aurita), we assessed the effects of sexual dimorphism, sample size, and validation methods on discrimination rates. We compared the three most common methods used to estimate the proportion of correctly classified males and females by DFA: resubstitution, jackknife, or sample splitting. Results from simulations indicate that these procedures may lead to opposite conclusions, especially when the sample size is small. In particular, the resubstitution techniques appear to be overoptimistic, and we therefore recommend that DFA accuracy be estimated by the jackknife cross-validation procedure. In addition, we show that most previous studies failed to present DFA accuracy with 95% confidence intervals, which hampers comparisons among studies. Finally, our results suggest that large sample sizes should be preferred over repeated measurements of the same individuals, because random measurement error is likely to have only a weak effect on the accuracy of the discriminant rate.