Summary: It is suggested that a convenient way of presenting the results of fatigue tests in which two different stress amplitudes are applied alternately is to plot log N against log (n1/n) where N is the total cycles to failure and (n1/n) is the fraction of cycles run at the high stress.With these co-ordinates, a simple geometrical construction gives a safe design method for the two-stress level system using only the conventional S-N curve and the value of (n1/n) expected to be encountered in service. If N1 and N2 are the lives at the high and low stresses as read from the S-N curve, one point may be plotted at (log N1, 0) since this represents the programme when all cycles are at the high stress. On the assumption, shown to be justified, that less than one cycle of high stress per 10,000 total cycles would not significantly affect the life at the low stress, a second point is plotted at (log N2, 4). The straight line joining these two points is always found to predict safe values of N for any value of (n1/n).This conclusion is checked against a wide range of experimental results taken from six different sources in the literature covering rotating-bending and push-pull tests, ferrous and non-ferrous metals, any order of stressing and length of programme cycle from 50 up to 5 million. This last feature means that the length of the programme cycle in service need not be known. All that is required is the proportion in which the two stress amplitudes are mixed. The average value of the ratio (experimental life/predicted life) for the data examined is 1·8, the extreme values being 1 and 56. By plotting in three dimensions an equation is also developed for the three-stress level spectrum and a suggestion is made for an extension of the method to multiple stress levels.