The eastern Baltic cod stock collapsed as a consequence of climate-driven adverse hydrographic conditions and over fishing and has remained at historically low levels. Spatio-temporal fishing closures [Marine Protected Areas (MPAs)] have been implemented since 1995, to protect and restore the spawning stock. However, no signs of recovery have been observed yet, either suggesting that MPAs are an inappropriate management measure or pointing towards suboptimal closure design. We used the spatially explicit fishery simulation model ISIS-Fish to evaluate proposed and implemented fishery closures, combining an age-structured population module with a multifleet exploitation module and a management module in a single model environment. The model is parameterized based on (i) the large amount of biological knowledge available for cod and (ii) an analysis of existing spatially disaggregated fishery data. As the population dynamics of eastern Baltic cod depend strongly on the climate-driven hydrographic regime, we considered two production regimes of the stock. MPAs were only effective for stock recovery when they reduced overall fishing effort. The performance of MPAs needs to be evaluated relative to environmental regimes, especially for stocks facing strong environmental variability.