Sweden reports large and variable numbers of human tularemia cases, but the high-risk regions are anecdotally defined and factors explaining annual variations are poorly understood. Here, high-risk regions were identified by spatial cluster analysis on disease surveillance data for 1984–2012. Negative binomial regression with five previously validated predictors (including predicted mosquito abundance and predictors based on local weather data) was used to model the annual number of tularemia cases within the high-risk regions. Seven high-risk regions were identified with annual incidences of 3·8–44 cases/100 000 inhabitants, accounting for 56·4% of the tularemia cases but only 9·3% of Sweden's population. For all high-risk regions, most cases occurred between July and September. The regression models explained the annual variation of tularemia cases within most high-risk regions and discriminated between years with and without outbreaks. In conclusion, tularemia in Sweden is concentrated in a few high-risk regions and shows high annual and seasonal variations. We present reproducible methods for identifying tularemia high-risk regions and modelling tularemia cases within these regions. The results may help health authorities to target populations at risk and lay the foundation for developing an early warning system for outbreaks.