Abstract: A method of real-time tracking a plane with maneuvering such as accelerating and turning, is proposed. It adopts a first order adaptive polynomial filter, in which the maneuvering is detected by using a X2-distributed statistics, that is, the square sum uN of N-predicted residues, and the variance of the model noise caused by maneuvering is estimated by two algorithms: Qk=Qk（uN/N- 1）/（M/N-1）, where M is a threshold for uN; and Qk=（xk）2. Numerical simulation shows that both are satisfactory for tracking a plane （see Figs, 1 and 2 ）, and the latter is more convenient for application when observations are not too bad.