To examine the long-term predictive power of heart rate variability (HRV) on all-cause mortality in randomly selected diabetic individuals. A total of 240 diabetic persons were randomly selected from the diabetic population. A 24-h ECG was obtained for each person included and analysed on the Pathfinder 700. In the RR Tools Program time (SDNN, SDANN, SDNN index, RMSSD, NN50, Triangular index) and frequency domain parameters (total power, VLF, LF, LFnorm, HF, HFnorm, HF/LF) were computed. After 15(1/2) years vital statistics were obtained. The analysis included 165 persons with acceptable ECG recordings. 81 individuals (49%) died during follow-up. Correcting for age and gender we found that in time domain, only the SDNN index was a significant mortality predictor but in the frequency domain, all parameters were significantly associated with death. In multivariate analysis only the power in the low frequency band was an independent predictor. During the period following the first 5 years, the baseline LF continued to be a significant predictor of mortality. This long-term follow-up study indicates that the LF power is the strongest HRV predictor with regard to mortality. A reduced HRV at baseline still holds prognostic information after 5 years.