This paper deals with the state filtering problem of nonlinear state-space functions with partial parameters unknown. A novel filtering algorithm is proposed based on a modified maximum likelihood method. Using a new likelihood probabilistic function, we obtain a two-stages strategy of Expectation and Maximization (EM) with the help of Monte Carlo method. The convergence of EM parameter updating strategy is verified theoretically and the proposed algorithm is an iterative process of these two stages. Here we use modified maximum likelihood method for parameter identification and particle filtering method for state estimation at current time moment. The proposed algorithm no longer uses all the observed values for iteration but only the current observed values, so it is simple for calculation and has advantages for application to online parameters identification. Simulation result proves that the proposed algorithm works effectively.