The life spans of animals can be measured in natural populations by uniquely marking individuals and then releasing them into the field. Selection on survival (a component of fitness) can subsequently be quantified by regressing the life spans of these marked individuals on their trait values. However, marked individuals are not always seen on every subsequent catching occasion, and for this reason, imperfect detection is considered a problem when estimating survival selection in natural populations. Capture–mark–recapture methods have been advocated as a powerful means to correct for imperfect detection. Here, we use simulated and field data sets to evaluate the effect of assuming perfect detection (‘naïve methods’), when detection is really imperfect. We compared the performance of the naïve methods with methods correcting for imperfect detection (mark–recapture methods, or MR). Although the effects of trait-dependent recapture probability are mitigated when recapture probability is high, mark–recapture methods still provide the safest choice when recapture probability might be trait-dependent. In our simulations, mark–recapture methods had a power advantage over naïve methods, but all methods lost statistical power at low recapture probabilities. The main advantage of mark–recapture methods over naïve methods is the ability to control for hidden trait-dependent recapture probability, as it is often hard to tell a priori if trait dependence is an issue in a particular study. However, when trait-dependent recapture probability is weak, naïve methods and mark–recapture methods perform similarly as long as recapture rates do not become too low, and the main problem of survival selection studies is still low statistical power. We provide a R package (EasyMARK) alongside with this paper to facilitate future integration between MR methods and classical selection studies. EasyMARK provides the opportunity to convert the regression coefficients from MR-approaches in to classical standardized selection gradients.