The stress relaxation data up to 8760h at 550℃ and 600℃of 12Cr-1Mo-1W-0.25V heat-resistant steel were used as the object to study the method of how to accurately and effectively predict long-term relaxation stress by using short-time relaxation data. When relaxation model is used to extrapolate the long-term relaxation stress directly, it is found that the parameters of the relaxation model depend on the length of the fitted data. The time-dependent parameter model, naming as timing parameter method, is proposed to predict the long-term relaxation stress with high accuracy. By comparison of the results of timing parameter method and direct extrapolation method, timing parameter method has obvious advantages in predicting long time relaxation stress with short time relaxation data, as the timing parameter method has a more accurate prediction than that of direct extrapolation method.