Bayesian state-space surplus production models are commonly applied in fisheries stock assessment when the only information available is an index of relative abundance. However, even relatively simple models such as these can be computationally expensive to fit, and diagnosing poor fits can be difficult. The Stan software package provides an advanced Markov chain Monte Carlo sampler and diagnostics that are not available in other packages for fitting Bayesian models. Here the sampler diagnostics, efficiency, and posterior inferences are compared among multiple parameterizations of a state-space biomass dynamics model, using both Pella-Tomlinson and Schaefer dynamics. Two parameterizations that prevent predictions of negative biomass are introduced, one of which allows for errors in catch. None of the parameterizations used avoid diagnostic warnings using the default sampler parameter values. Choosing the appropriate parameterization of a model, and paying attention to these diagnostics can increase computational efficiency and make inferences more robust.