The fundamental problem of nonlinear filtering theory is how to solve robust D-M-Z equation in real time and in memoryless manner. This paper describes a new real time algorithm which reduces the nonlinear filtering problem to off-line computations. Our algorithm gives convergent solutions in both pointwise sense and L/sup 2/ in case that the drift term and observation dynamic term have linear growths. The algorithm presented is slightly better than that given in our previous paper (2000).