At the present time, it is generally conceded that testing for the evaluation of fatigue performance of materials subjected to variable loading in service, must contain variable loading as well. Single-level fatigue testing just will not provide an accurate answer.Two methods of fatigue testing using variable amplitude have been advocated in the literature. The earlier method, developed by Gassner, involves the stratification of the service load spectrum into eight levels, from the maximum load to the minimum. The number of cycles at each level is developed from the load spectrum. The loads are then applied in a stepwise ascending-descending manner beginning with a mid-level.The second method involves the use of a clipped random process, whose intensity (RMS) is varied throughout the test in a random stepwise manner, such that the final accumulation of loads is asymptotic to the desired load spectrum.These two methods are examined in detail in this paper. A test program, involving automotive stub axles was developed, in which as many factors as possible were kept common to both types of tests. It is hoped that the results are a fair assessment of the two methods.