Abstract: By using viscosity testing data of Tahe Heavy Oils mixed with 4 kinds of Tahe Light Oils, Arrhenius viscosity model, double logarithmic viscosity model, Cragoe viscosity model and Lederer viscosity model are evaluated. The results show that Cragoe model has the highest accuracy and is most suitable for prediction of the viscosity of Tahe Heavy Oil mixed with light oil. In consideration of impacts of different mixed light oil volumes on prediction error of the Cragoe model, mass fraction of the heavy oil is taken as an improvement parameter of the Cragoe viscosity model to conduct increase and reduction of predicted values respectively for light oil-heavy oil volume ratios higher and lower than 1:1, so that the Cragoe model is improved to reduce average error and maximum error respectively by 43% and 32%.