Nightjars have in Britain traditionally bred on heathland, where their numbers have declined slowly but steadily since the beginning of the century. In recent years, however, the numbers on recently replanted forestry sites have increased dramatically. There are differences in density and behaviour in the two types of habitat which have been followed to some extent by ringing and radio tracking. Both of these techniques have limitations, and recognition of individual males by song could make a useful contribution to the study of territories and movements during and possibly between seasons. A small population of nightjars exists on Skipwith Common in North Yorkshire and has been fairly stable at between 2 and 8 for the last twenty years. The Common has regions of dry heath interspersed with scrub, mature oak woodland, several areas of wet heath and a considerable number of pools. Active management to clear scrub and restore dry heath regions started in 1993. Nightjar territories are well dispersed. This study followed all males singing on the Common between 13th June and 13th August. A single bird singing in a forestry plantation about 8 km away was also recorded. Territories were mapped using observations of successive songs. Where possible this was done on several different nights for each bird. The order of use of songposts differed but the outlines remained similar until the end of the singing period. When simultaneous singing with a neighbouring bird could be heard this was also noted, but the density was not sufficiently high for this to be the principal mapping technique and could only be used to confirm parts of territory boundaries. With territories established in this way, songs of each churring male were recorded using a Sennheiser ME62 microphone in a parabolic reflector. Recordings were made in poor weather as well as fine, at dawn as well as dusk and over distances between 20 and 200 m. The Common is surrounded by coal mines, which, with traffic noise, contributed considerable background noise. The recorded songs were digitised using an SB 16 sound card and analysed using the associated WaveStudio software and a custom built sonogram program which includes simple statistical tools. In all, sixteen variables were considered, six frequency-based, two power-based and eight time-based. Of these, two time-based variables, the rate of churr in the high- and low-pitched parts of the song, proved the most robust. These two variables alone made it possible to distinguish between the songs of the small population at Skipwith Common. Results are presented for these and two other birds recorded in the same season, showing the effect of all extremes of recording conditions. The robustness and usefulness of the other variables is assessed and suggestions for choice of variables with larger populations made.