A model for on-line calculation of equivalent stresses in steam turbine rotors has been built up based on Support Vector Machine (SVM) neural network and the finite element method. Regarding 125MW steam turbine rotor as an object, SVM method is employed as a tool to train large numbers of date about the equivalent stress of the rotor, which obtained from the finite element method. A model can be gotten after finishing training the data. Equivalent stresses of rotor obtained with finite element and SVM model are compared during cold start. Results show that similar calculation results of equivalent stress may be obtained with either finite element or SVM model. However, with the latter one, the calculation becomes easier and fast, which is applicable for on-line calculation of equivalent stresses in turbine rotors, and may serve as a reference for on-line life management of turbine rotors.