In cross-directional (CD) control processes,the system model is difficult to identify due to the sparse measurement data as well as the model 's high dimensionality,strong coupling and uncertainty.In order to identify the system model rapidly and accurately,an improved approach on the basis of the recursive Levenberg-Marquardt (RLM) algorithm is proposed.Firstly,the two-dimension parameterized model and step identification method of CD control processes are described.Secondly,the objective function of RLM algorithm is modified and an improved RLM algorithm is presented,which is then applied to the real-time and recursive identification of the mapping,steady-state spatial response and dynamic response of the parameterized model of CD control processes.The results of both simulation and practical application show that the proposed identification method can not only identify the spatial and dynamic responses of CD control processes simultaneously but also converge more rapidly than the traditional RLM algorithm.