A practical control method using neural network and fuzzy control techniques is applied to the level control of the tank in a solvent dewaxing plant. The purposes of the control are to stabilize the tank level and to smoothly change the outflow rate from the tank which conflicts with the stabilization of the tank level. This plant has non-linear characteristics such as refrigerating process and filtering process, so it is difficult to predict the inflow rate to the tank. Especially, the response of the inflow rate to the tank has a large dead time when the feed oil to the plant is switched. Furthermore, the heater installed in the down stream of the tank restricts the outflow rate from the tank. This neuro-fuzzy hybrid control system consists of three components, (1) a statistical component, which calculates long-time tendencies of the outflow rate from operation data. (2) a correction component, which is a fuzzy controller for compensating the outflow rate calculated by the statistical component, and (3) a prediction component, which uses neural networks to generate suitable control target patterns for the fuzzy controller in advance. This neuro-fuzzy hybrid controller for the real plant works well not only in steady state but also in transient state.