In order to find appropriate experimental conditions for the study of low-cycle random fatigue of steel, a numerical experiment with a model of the fatigue mechanism was carried out. The procedure was almost the same as was done in the second report except that statistical properties of materials were taken into account. Rotating bend tests under random loading were carried out to compare with the numerical experiments. Conclusions obtained were as follows, 1) In order to simulate random fatigue behavior, it is necessary to take statistical properties of materials into consideration. Generally speaking the effect of randomness of applied load is greater than that of variations in materials properties when R.M.S. is large. 2) Both mean value and coefficient of variation of fatigue life tend to increase as the correlation in applied random load increases. Therefore correlated random load sequences must be used in experimental studies of random fatigue. 3) Both numerical experiments and rotating bend tests showed that the effects of constant load block size (DN) is remarkable for correlated load sequences. DN must be made as small as possible, say less than 25 cycles for the experiments where experimental conditions are similar to the present study.