ABSTRACT We examine the problem of estimating the mass range corresponding to the observed red supergiant (RSG) progenitors of Type IIP supernovae. Using Monte Carlo simulations designed to reproduce the properties of the observations, we find that the approach of Davies & Beasor significantly overestimates the maximum mass, yielding an upper limit of Mh/M⊙ = 20.5 ± 2.6 for an input population with Mh/M⊙ = 18. Our preferred Bayesian approach does better, with Mh/M⊙ = 18.6 ± 2.1 for the same input populations, but also tends to overestimate Mh. For the actual progenitor sample and a Salpeter initial mass function, we find $M_{\rm h}/\mathrm{M}_\odot = 19.01_{-2.04}^{+4.04}$ for the Eldridge & Tout mass–luminosity relation used by Smartt and Davies & Beasor, and $M_{\rm h}/\mathrm{M}_\odot = 21.28_{-2.28}^{+4.52}$ for the Sukhbold, Woosley & Heger mass–luminosity relation. Based on the Monte Carlo simulations, we estimate that these are overestimated by $(3.3\pm 0.8)\, \mathrm{M}_\odot$. The red supergiant problem remains.