probabilistic assessment is made of the design S-N curves for 1Cr18Ni9Ti pipe-welded joint. The material exhibits a character of random cyclic stress-strain and fatigue life responses. Then, both the S and N in the virtual stress amplitude-fatigue crack initiation life (S-N) data are random variables. A general maximum likelihood method, which considers the randomness of the test S-N data, is developed to model the probabilistic S-N relations of material. The design S-N curves with reduction factors of 20 on cycles, 2 on stress and 2.5 on stress applied to the best-fit S-N curve of test data as used in the ASME code and general practice, are assessed in a probabilistic sense. The results reveal that in the lower S regime the 20 and 2.5 reduction factor curves may be appropriate, while the reduction factor of 2 is slightly nonconservative. But in the higher S regime these curves are all much more conservative. The intrinsic cause is that the standard deviation of logN exhibits an increase with the S decreasing, while the constant reduction factors do not consider this increase. It is shown that from the general statistical scatter of S-N data the design S-N curves should be appropriately determined by a probabilistic approach.