In case goodness-of-fit of some assumed models which represent one phenomenon has to be tested, several goodness-of-fit significance tests are used. Each one of them can give the order of goodness-of-fit among assumed models relatively, but not absolutely. So it is very difficult to determine which test must be used in such cases. The purpose of this paper is to discuss the accuracy of goodness-of-fit significance tests which are usually used in testing the goodness-of-fit of traffic assignment models-i.e. CHI-square value, correlation coefficient, root-mean-square error, and weighted-root-mean-square error. The following results are obtained under the condition that the number of traffic routes is small : (1) correlation coefficient is not so good as others; (2) the difference in accuracy among CHI-square value, root-mean-square error, and weighted-root-mean-square error is not significant.