Several kinds of regression functions of SN curve for fatigue test have been suggested and the least squares method is used to determine parameters in each equation. In the case of a non-linear function, however, the application of the least squares method cannot always enough explain properties of objective lots. In this paper it has been clarified that the application of the experimental regression analysis is able to correct these defects and to estimate the parameters in the functions and regression error. In the experiment the lots of samples consist of specimen which have been manufactured considering materials, conditions of heattreatment, hardness of specimen, and so on. Fatigue strength (or SN curve) of each lot and overall estimated error are obtained by using the method of experimental regression analysis. The estimated error of the regression functions in this experiment is 21. 6-31. 4 MPa in standard deviation. As the application method of this new system in future, it is indicated that new obtained data of fatigue test in addition to the group of accumulated data are analyzed and that SN curves are able to be estimated consistently each other even when a number of new data is small.