In this paper, an incremental 3DVAR data assimilation scheme for the ARPS model has been developed in which a cost function is defined as a sum of the background field constraint and the observation constraint. The background field can be provided by a single sounding, a previous ARPS forecast, or another operational forecast model, such as the RUC operational numerical model from NCEP. Observations used include: single-level surface data (such as Oklahoma Mesonet), multiple-level or upper-air observations (such as rawinsondes and wind profilers), as well as Doppler radar observations. The background error covariance matrix is modeled by a recursive filter and is used as a preconditioning. This preconditioning prevents the smallest eigenvalue of Hessian matrix of cost function from becoming less than one and improves the convergence rate of the minimization. Some numerical experiments have been conducted based on this scheme. It is shown that the quality of the assimilation, which analyzes the initial conditions based on both observational and background information, is reasonable. Single observation experiments show that the recursive filter performs adequately in spreading the observational information. Also, the strength of its effect depends on how many passes are used and on the influence radius of this filter.