The structures of railway vehicles are required to service in super-long life regime. Determination of the probabilistic S-N curves including the regime should be a basic work to realize the real fatigue life prediction and reliability assessment. Based on the test results of Chinese railway LZ50 axle carbon steel, a statistical extrapolating method is proposed to determine the curves by applying the conventional test data in mid-long life regime. Some phenomena, i.e. the response of “fatigue limit”, a great of heat affecting using the high frequency ultrasonic fatigue test system, and the non-conservative test results using the conventional frequency tiny multi-specimen test system (machine C), are firstly mentioned in the existent super-long life researches. The too much conservative offers are also noted by the determinations of existent codes. To address the variable amplitude loads in production, the present method extrapolates continuously the curves from the mid- to super-long life regimes under the statistical controls of conventional fatigue limits. The practice for the present material indicates that the curves can well reflect the scattered data not only in mid-long life regime but also in super-long life regime. The non-conservative test results using machine C are also safely included by the curves.