Management advice for hard-to-age species such as prawns, crabs and rock lobsters are usually based on size-structured population dynamics models. These models require a size–transition matrix that specifies the probabilities of growing from one size-class to the others. Many methods exist to estimate size–transition matrices using tag–recapture data. However, they have not been compared in a systematic way. Eight of these methods are compared using Monte Carlo simulations parameterised using the data for the tiger prawn (Penaeus semisulcatus). Four of the methods are then applied to tag–recapture data for three prawn species in Australia’s Northern Prawn Fishery to highlight the considerable sensitivity of model outputs to the method for estimating the size–transition matrix. The simulations show that not all methods perform equally well and that some methods are extremely poor. The ‘best’ methods, as identified in the simulations, are those that allow for individual variability in the parameters of the growth curve as well as the age-at-release. A method that assumes that l_∞ rather than k varies among individuals tends to be more robust to violations of model assumptions.