Fuzzy models with hybrid algorithms of adaptation allow to predict complex temporary processes successfully. This article shows how we can use fuzzy difference TSK model for traffic flow predictions. Described hybrid algorithm, it uses fuzzy logic, neuro - fuzzy network and recurrent algorithm. Hybrid algorithm consists of two types identifications: parametric and structural. We apply it for traffic data from California and compare with another prediction methods.