This paper discusses the margins of the design fatigue curve in the ASME Boiler and Pressure Vessel Codes Section III from a reliability analysis point of view. It is reported that these margins were developed so as to cover uncertainties of fatigue data scatter, size effect, and surface condition[1], but the reasons for them remain unclear. In order to investigate the physical implications of the design margin, a probabilistic approach is taken for the collected fatigue data of carbon and low-alloy steels. In this approach, these three parameters are treated as random variables, and an applied stress is also taken into consideration as a random variable. For the analysis, to begin with, a limit state function for fatigue is proposed. Next, reliability index contours of the design fatigue curves for carbon and low-alloy steels are obtained based on the proposed limit state function. The contours indicate that the margins 2 on stress and 20 on life do not provide equal reliability. The margin 20 on life is more conservative and the margin became a minimum near intersections of the design curves with margins 2 on stress and 20 on life. For practical applications, the partial safety factors (PSF) for the target reliability are computed for all materials and several levels of coefficients of variation (COV) of the applied stress. A sensitivity analysis of the PSFs clarifies that only two parameters, the strength (or the life) and the applied stress, are predominant. Thus, the partial safety factors for these two parameters are proposed in a tabular form.