The problem of approximate realisation is described and various methods are discussed. A new method is then given which directly identifies the system poles from pulse response data by finding the local minima of a real function of a complex variable. These estimates of the poles are then refined using a new iterative nonlinear least-squares algorithm. Finally, these methods are applied to a ‘seismic wavelet’, and are shown to give good qualitative and quantitative information on the system being modelled.