Recordings of elephant vocalisations can be used to determine the size and composition of the herd, the sexual state, as well as the emotional condition of an elephant. Manual analysis of recordings (by listening to these and by visual inspection of spectrograms) to locate vocalisations is tedious. The automatic detection of vocalisations in recordings is explored. Important signal characteristics of elephant vocalisations were identified from spectrograms and a technique, based on the principles of existing voice activity detection algorithms, was developed to exploit these. Results obtained suggest that the algorithm can reliably detect elephant vocalisations from noisy recordings as long as the harmonic structure of vocalisations is not buried in background noise.