Three-stage algorithm of parameter estimation for autoregressivemoving average(ARMA)models with coloured observation noise is presented.In the first stage,the autoregressive(AR)parameter estimation is given byusing the recursive extended least squares(RELS)method. In the secondstage,the parameter estimation of adjoint moving average(MA)model is giv-en by using the Gevers-Wouters iteration algorithm. In the third stage, theoriginal MA parameter estimation is given via solving a inconsistent set of Lin-ear equations by using the least squares method. A simulation example showsits effectiveness.