A variant of Ensemble Square Root Filter (EnSRF) referred as iterative EnSRF is designed according to asynchronous algorithm. The performance of iterative EnSRF is examined using Lorenz96 model. Unlike traditional EnSRF, the iterative EnSRF can synchronously update two model states at different time and improve the analysis by iterative procedure. The performance of iterative EnSRF is examined not only by using different kinds of observations but also by using perfect and imperfect models. Meanwhile, the rationality of iterative EnSRF analysis is also discussed. With a perfect model, iterative EnSRF is able to increase the convergence speed of regular data assimilation and analyze the indirect observation more effectively. With an imperfect model, iterative EnSRF cannot effectively improve the analysis for all tested observations. If the incorrect parameter is perturbed, iterative EnSRF is able to utilize the improvement of ensemble forecast system to optimize the analysis for parts of observations. Further investigation of experiment results indicates that the improvement of iterative EnSRF analysis is contributed to the optimization of spatial structure of background error covariance and the linear assumption of EnSRF being more reasonable in iterative EnSRF procedure.