Fed-batch cultivation of Azotobacter vinelandii 21was optimized for biosurfactant production. Optimization of feed-rate time profile and concentration of nutrient medium components in feeding solution is based on a hybrid mathematical model consisting of mass-balance equations for biomass, biosurfactant, volume of cultural liquid and substrate components: glucose, ammonia nitrogen and phosphate phosphorus. The rate of cultural liquid emulsification activity growth as well as the rates of ammonia nitrogen and phosphate phosphorus consumption is modelled by means of artificial neural network, while the rates of the other biochemical transformations are modelled by adequate kinetic relationships.