The purpose of this study is to investigate the effects of IRT linking methods when the average ability of examinees for the new test is significantly higher than the average ability of the original norm. the IRT linking methods included are “mean and sigma”, “iterative mean and sigma”, “iterative robust mean and sigma” and “item characteristic curve”. Four linking settings for the target scale are (1) true values for the item and person parameters, (2) true values for the item parameters but estimated values for the person parameters, (3) estimated values for the item parameter but true values for the person parameters, and (4) estimated values for the item and person parameters. The true values of the item parameters (both for the anchor tests and the whole tests) are used as the criteria for evaluation. Three replications are implemented and the average results are presented. The evaluation of the linking effect is clearer and more optimistic than the results form the previous studies. No matter whether the parameters of the target scale are true values of estimated values, the inferences are in agreement. In other word, the inference made from the real world’s data will be valid for the true value. For both the difficulty, b, parameter and the discrimination, a, parameter, the best choice for linking is the method of “item characteristic curve” and “mean and sigma” depending on the criterion adopted. If judging form the whole test, the method of “item characteristic curve” is be the best choice. When the difficulty parameter of the anchor test it the method of “mean and sigma” performs best and the method of “item characteristic curve” is the second choice. Generally speaking, the method of “item characteristic curve” is recommended for the data sets simulated. Especially for the discrimination parameter, the method of “item characteristic curve” performs best consistently. The method of “mean and sigma” is recommend for its computation simplicity. Linking process decreases the scale difference which was caused by the different samples of examinees. Linking error is minor compared with the estimation error, especially for the difficulty parameter.