In this paper the notion of spectral interactor matrix (SIM) for a given square transfer matrix is introduced and used in the singular filtering problem. A SIM is defined as an interactor matrix preserving the spectral properties of the underlying system. After studying the class of all spectral interactor matrices associated with a given transfer matrix, an algorithm for the computation of a SIM is proposed. By resorting to the SIM, it is possible to point out the role played by delays (zeros at infinity) and nonminimum phase zeros in the solution of the singular filtering problem.