This paper presents a new design method for optimum clutter IIR filters when projection initialization is used. This method minimizes the mean squared-error between the steady-state and the dynamic frequency responses. The designed filters offer better dynamic frequency responses for a given number of samples when compared to steady-state filters without initialization or with projection initialized filters. The improvement is achieved by redesigning the filter's feedback coefficients to minimize the mean squared-error for a given number of samples. New clutter filters were successfully designed for processing different numbers of samples and their performances were studied by comparing their dynamic frequency responses with the desired steady-state frequency responses.