A computerised method for the analysis of QT intervals in ambulatory ECG recordings is presented. This approach is based on selective beat averaging which allows one to process P-QRS-T complexes together with the environment that characterises them. Long-term autonomic nervous system influences are accounted for by separating the analysis over different circadian periods. Effects of QT recovery time are taken into account by requiring a stable heart rate preceding each beat to be averaged. Before averaging, beats are resampled and realigned with respect to the R-wave peak estimated by parabolic interpolation. Averaged ECG templates are then analysed with an algorithm which automatically detects QRS complex and T-wave features. Repolarisation analysis is based on first and second derivatives of lowpass filtered ECG (recursive Butterworth filter). The QT/RR relationship and the circadian QT variation at identical heart rate were analysed in 14 normal individuals. When performed at stable heart rate conditions and when confined to well-defined circadian periods, the QT/RR relationship was strongly linear (r = 0.95 +/- 0.06); in addition, the slope of this relation changed between day and night (respectively, 0.197 +/- 0.07 and 0.139 +/- 0.03, p < 0.01). The range of circadian QT variation at identical heart rate was approximately 20 ms for both males and females.