In this paper drifting nature in the evaluation conclusion of single method is investigated in order to offer the experiment basis for combination evaluation. First, real evaluation problems are divided into 3 classes and emulation experiments are carried on them separately. Then, objects to be evaluated are generated by Matlab 6.1, and are evaluated by using different methods. Third, the compatibility of the comprehensive evaluation methods is discriminated. Finally, the average of the compatible methods is adopted as reference frame in order to measure drifting degree of each method quantitatively.