We present a risk assessment perspective on earthquake recurrence on mature faults and the ways that it can be modeled. Specific needs unique to risk models compared with probabilistic seismic‐hazard assessment include (1) the determination of loss per event to assess the correlation of risk between sites for portfolio management, (2) loss at all return periods, and sometimes (3) realizations of the order in which events occur. We use three mature faults in New Zealand as a case study and review the impact of parameter and model uncertainty on risk metrics. We review the physical characteristics of several faulting environments, contrasting them against the properties of three probability density functions (PDFs) widely used to characterize such interevent time distributions. We review the data available to constrain both the priors and the recurrence process and conclude that to use them to quantify the recurrence of large events on mature faults, it is best to compute a Bayesian combination of models (i.e., the weighted average of individual PDFs using their Bayes factors). Finally, we propose the following to the community: (1) the initiation of a general discussion on how best to incorporate our knowledge (e.g., from earthquake geology) on plausible models and model parameters, while preserving the information on what we do not know and (2) the creation and maintenance of a global database of priors, data, and model evidence, classified by tectonic region, fluid characteristic, fault geometry, and other relevant properties so that we can monitor whether the time‐dependent recurrence behavior of certain types of faults or faulting environments can be better described by one (or a combination of) recurrence model(s).