Abstract A stochastic semi-annual model with an autoregressive stock–recruitment component was used to evaluate stock dynamics of North Sea sandeel ( Ammodytes marinus ). Parameter estimates were obtained with Markov Chain Monte Carlo and maximum likelihood methods. Historical estimates of spawning stock biomass (SSB) were determined with an average coefficient of variation of 22%. Status quo predictions showed that it was possible to predict SSB 1 year ahead with about the same precision as the average. In predictions further ahead the CV increased to between 50 and 70% mainly due to the high uncertainty inherent in the recruitment predictions. Adult sandeels are stationary and based on results from larval drift models it has been suggested to divide the North Sea into smaller assessment areas. However, contrary to expectations, separate regional assessments resulted in one case in even more uncertain SSB estimates than the overall North Sea assessment.