The design flood is defined as the discharge value corresponding to an assigned non-exceedance probability, which is usually expressed in terms of the return period. Estimation of the design flood is usually carried out by fitting observed data samples with a suitable probability distribution. The objective of this study is to evaluate if model selection criteria, which are seldom used in hydrological applications, can help identifying the best probability model for this purpose. The study analyzes the performance of three model selection criteria, namely, the Akaike information criterion, the Bayesian information criterion, and the Anderson–Darling criterion. The three methods are compared trough an extensive numerical analysis by using synthetic data samples. The study demonstrates that model selection criteria are a valuable tool for reducing the uncertainty of design flood estimation.