A simple method of nonlinear filtering is applied to the problem of dynamic reactivity estimation in which the law of reactivity change is assumed to be unknown. The filter is designed based on a system model containing the usual point reactor kinetics equations driven by fictitious white noises and a reactivity state equation. The latter is formulated such that the rate of the reactivity change is a random process, taking account of the unknown reactivity change. The nonlinear filter applied here is a simple modification of the Kalman filter added with a nonlinear feedback loop. The key parameter that determines the filter response is the parameter of the fictitious noise in the reactivity equation which is closely related to the filter gain. The results of the computer simulation and the experiment show that the nonlinear filter can be used to estimate the dynamic reactivity, even under an extremely noisy measurement condition.