Arrays of seismometers, spaced over a distance comparable to the longest apparent wavelength of the signal, have been used to facilitate the separation and identification of seismic phases by a process of velocity filtering. Individual seismometers in the array are recorded on separate tracks of magnetic tape, and summed on playback with the insertion of time delays to cancel the propagation delays of the required signal. The inserted delays have the effect of tuning the array to a required velocity and azimuth, thereby discriminating against signals or noise of different velocity or azimuth. The improvement in signal/noise ratio of weak signal onsets enables a more accurate estimate to be made of epicentre location. Correlation methods, presenting the correlator output as a function of velocity or azimuth, are an objective aid to identifying seismic phases. Velocity filtered records of the Gnome nuclear event, recorded by an experimental array at a distance of 1000 km, enabled at least ten seismic phases to be identified, many of them being multiple reflexions from the Mohorovicic discontinuity. The travel times of the reflexions enabled the mean thickness of the crust along the propagation path to be deduced. Examples are given to show the application of array techniques to teleseismic events, including the detection and identification of core phases, and reduction of noise generated by the signal in the vicinity of the recording station. The theoretical responses of arrays of several configurations are presented, and their relative merits discussed. The theoretical performance of cross arrays is given in detail, including the application of correlation methods and a discussion of errors in velocity or azimuth determination. Theoretical expressions are derived for correlator signal/noise ratios in the case of random noise. Details are given of the experimental arrays, including siting criteria. The present analogue processing facility, and a special purpose computer under construction, are also described.