In this paper, new neural network approach for characterization of cylindrical metallic cavity loaded with multilayer dielectric is suggested. Such cavity configuration has a great importance in realization of microwave resonant applicators, widely used for thermal processing of multilayer materials. Proposed neural model is based on representation of multilayer cavity load via several planparallel homogeneous dielectric slabs and superposition of separate influence of each dielectric slab parameters on the cavity resonant frequency. Model is incorporated into MLP neural network enabling its efficient implementation on modest hardware platform. Suggested approach is verified on the example of circular cylindrical metallic cavity loaded with two-layer dielectric.