Solution of stochastic nonlinear identification problem is proposed. An algorithm of discrete observer parameter identification, based on utilization of criterion of minimal estimation error probability, is synthesized. To illustrate proposed approach efficiency we consider a model example.