Multivariable nonlinear dynamic processes can be generally described as nonlinear ARMAX models. In this paper, we present a new algorithm for model structure determination and parameter estimation of systems of this sort. The structure and parameters can be obtained simultaneously and recursively. It is based on the "innovation" idea and certain information criteria. Using the technique of matrix operation, it is possible to combine innovation calculation with structure determination and parameter estimation. The amount of computations is thus reduced considerably. This procedure assures that the optimal system description is obtained. The identification of a simulated system show the correctness and effectiveness of this new algorithm.