Gas welding is a very important and useful technology in the fabrication of railroad cars and commercial vehicle structures. However, since the fatigue strength of gas-welded joints is considerably lower than that of the parent material due to stress concentration at the weld, the fatigue-strength assessment of gas-welded joints is very important for the reliability and durability of railroad cars and the establishment of a criterion for long-life fatigue design. In this paper, in order to save time and cost for the fatigue design, an accelerated life-prediction method that is based on the theory of statistical reliability was investigated. Its usefulness was verified by comparing the (Δσa)R-Nf relationship that was obtained from actual fatigue test results with the (Δσa)R-(Nf)ALP relationship that was derived from accelerated life testing. And the reliability of the predicted life was evaluated. The reliability of the accelerated life-prediction on the base of actual test data was analyzed to be (81～86)% of the actual test life of the fillet-type gas welded joint.