It is commonly assumed in analysing fatigue data that there is a definite functional relationship between life in number of cycles and stress level. However, as has been pointed out several times (1, 2), an examination of the data shows considerable scatter. Even with carefully prepared smooth specimens, all from the same heat of steel, treated in the same manner and tested in the same laboratory, a range of 2 to 1 in number of cycles for failure at the same stress level is normal (1) and a range of 10 to 1 is not unusual (2). If the specimens are tested by different laboratories, slightly varying techniques will introduce further scatter (3).