Abstract: A neural network based adaptive controller for a multivariable fermentation process is proposed,in which the linearizing method of the nonlinear systems is combined with the on line identifying technology of the neural networks.When adequate prior information concerning the dynamics of the fermentation processes is not available and parameters of the processes changes with time,the simulation experiments demonstrate that the presented approach proves more effective than the input/output linearization method based on differential geometry.Moreover,because in the input/output linearization method the design of the controller relies on the exact process model and its control performance is sensitive to the parameters of the model,the method results in low accuracy and poor robust.