Statistical office of the republic of Slovenia (SURS) currently publishes annual survey unemployment estimates on NUTS3 level. These annual estimates are often judged as less reliable (CV>10 %). We successfully applied small area estimation techniques to quarterly and yearly NUTS3 estimaties of survey unemployment and survey activity rates. Bayesian hierarchical estimates proved significantly more precise compared to direct estimates. We ascribe this improvement to the inclusion of rich variety of auxiliary data on personal and area level.