A limiting factor in the design of any Digital Subscriber Loop system will be the amount of impulse noise on subscriber lines. In this paper a new method for detecting impulse noise events using a statistical measure is presented. This is described in comparison with an existing technique which uses a hard amplitude-time threshold to locate the impulses. Both techniques are then used to calculate impulse duration and amplitude PDFs from real subscriber line data, recorded in the UK. Next these PDFs are compared to an analytical model proposed by Deutsche Telekom (DT). It is found that the form of the model is correct, but that the parameters chosen by DT are not appropriate for the line data used in this study, which was obtained from the network of British Telecom. Finally, more suitable parameter values are suggested in order to arrive at a working model of impulse length and amplitudes.