The problem of rapid detection of an unexpected reactivity insertion into a nuclear reactor is studied assuming a stochastic point reactor model and noisy measurements of neutron density. The fundamental assumption is that the time dependence of the reactivity is given as in a ramp function with unknown coefficients. Thereupon, the present method applies a likelihood ratio test to the innovation sequence obtained by using a discrete Kalman filter, which is designed for the steady-state condition of reactor operation. By numerical experiment, the mean delay time for detection has been obtained under the condition that the mean time between false alarms takes on a prescribed constant. A comparative study with some typical existing methods shows that the proposed method is remarkably effective except for extremely large or small inputs of reactivity.