In this paper, in order to save time and cost for the fatigue design and to develop the optimum approaches for accelerated life prediction of the fillet gas welded joints, the $({Delta}{sigma})_R$ - $N_f$ relationship was obtained from actual fatigue test data, including welding residual stress. Based on these results, the $({Delta}{sigma}_a)_R$ - $(N_f)_{ALP}$ relationship derived from the method of statistical probability analysis was compared with the actual fatigue test data. From the result, the optimum statistical distribution for the accelerated life prediction was analyzed to be the lognormal distribution for the fillet-type, gas-welded joint. The mean accuracy of the accelerated life prediction was assessed to be 85~95% of the actual test life at the 95% reliability level and ${pm}15%$ standard deviation. Therefore, it is expected that the accelerated life prediction will provide a useful method for determining the criterion for fatigue design and for predicting a specific target life.