Fisheries management strategy evaluation ranks candidate management procedures according to their relative risk of failing to meet management outcomes, examining distributions of possible futures generated by monte-carlo simulations. The historical period, which sets up the jumping off point of these simulated futures, is often conditioned on the results of an assessment, usually by sampling the posterior distribution of the assessment in some way to integrate over uncertainty. Previous research on best practices suggests that sampling a Bayesian posterior is the best way, but makes no mention of how exactly the sampling should be designed. Furthermore, while a small number of posterior samples may be adequate to rank management procedures, objective risk measures may be biased under small sample sizes. We show two examples of sampling procedures that we used to improve representation of risk in previous management strategy evaluations for BC herring and sablefish, and discuss preliminary results from rigorous tests of alternative sampling designs.