The problem of separating unknown signals from much stronger ‘coloured’ noise often arises in radar receivers (m.t.i. cancellers) and antenna sytems (directive noise in phased-array antennas). Two main optimisation criteria can be used to design rejection filters, namely (i) the maximisation of average signal/interference ratio (based on the evaluation of the eigenvectors of the interference covariance matrix) and (ii) the linear prediction of interference samples (based on the solution of the normal-equations set). These two methods are compared in terms of algorithms complexity and improvement of signal/interference ratio. The results of computer evaluation show that in most practical cases, i.e. when the interference has a correlation coefficient greater than 0.8, the performance of the suboptimal method of linear prediction is very close to the optimum one. Moreover, the linear prediction method gives a greater flatness in frequency response and, as far as implementation is concerned, is much simpler.