Abstract: The paper introduces a kind of adaptive neural-fuzzy inference systems（ANFIS） based on T-S model to deal with the modeling problem of the complex soda carbonization process.With the measured temperature and soda-output flux data，the system self-initializes the fuzzy rules，and adjusts the membership parameters independently until to find the nonlinear flux-temperature model of the process.The paper discusses the system’s structure and its learning algorithm.Simulation results demonstrate the good practical application value of this approach.