Population models form the basis for the assessments of species in the tiger prawn component of Australia's northern prawn fishery. Penaeus semisulcatus and P. esculentus are assessed using a size-structured population model. These assessments form the basis for a control rule which predicts future total allowable catches (TACs) for P. semisulcatus and P. esculentus so that the discounted profit from the fishery is maximized. However, there are concerns with this approach: (i) the TAC predictions have consistently overpredicted actual catches and (ii) the assessment for one of the species exhibits a retrospective pattern. A series of analyses was conducted to explore the causes of these observations. Results indicate that catch, effort, and recruitment prediction can be improved substantially by changing the assumed selectivity pattern for one of the surveys, changing how the length frequency data are assembled from the raw data collected, changing the constraints on the minimum amount of effort by target fleet, modifying how the distribution of effort by week is forecasted, and dropping the length frequency data from the most recent recruitment survey. More generally, the analyses illustrate how retrospective analysis can be used to improve how assessments and projections are undertaken when the quantities of interest are known retrospectively.