The ultrasound technique (20 kHz loading freqeuncy) has been investigated as a time- and energy-saving method for measuring lifetimes under service loading conditions. AISI C1020 steel was tested between 3 x 106 and 3 x 109 cycles with two Gaussian-like random loading programs (Gauss distribution generated with a Markov matrix and straight-line distribution). The resulting lifetime curves can be approximated by straight lines in a log-log plot. If the maximum values are plotted, these lines lie above the S/N curves and have approximately the same slope as the S/N curve for finite lifetimes. The experimentally found lifetimes are compared with predictions according to Miner and the Miner-Haibach rule (with half the slope of the S/N curve in the endurance range). Good agreement is found for measured and calculated results according to the Miner-Haibach rule if the measured amplitude distributiom is introduced into the calculations. This agreement is especially good for Markov random loading. Predictions according to the original Miner's rule give lifetimes that are too long in the very high cycle range. This result is explained by damaging effects of amplitudes below the endurance limit. For linear distribution random loading this effect has not been observed in the cycle range up to 3 x 109.