The purpose of this study is to examine the possible effects of the four IRTlinking methods of the three subtests of elementary school Quantative AptitudeTest when across grades' comparison is necessary. The three-parameter-modellinking methods adopted were mean and sigma, robust mean and sigma, iterativerobust mean and sigma, and item characteristic curve transformation. The subtestsincluded were Quantative Relations, Quantative Commutation and Number Series.For each subtest, 15 anchor items were used for all fourth, fifth and sixthgraders. The estimated item parameters of the sixth graders were used as theevaluation criterion of the linking effects. The correlations between the fifth andsixth graders were around .85, but dropped to .65 for fourth and sixth gradersfor the difficulty parameter of the anchor items. The correlations for thediscrimination parameter were worse (around .63 and .10 respectively for thefifth and fourth graders). The linking results confirmed the concern of the linearassumption. Generally speaking, method of mean and sigma is suggested especiallywhen the the discrimination parameter can not be appropriately estimated. Thecomparison between two continuous grades is probably the most we can tolerantfor any linking method to be able to adjust the scale differences. The verticallinking across three grades is not recommended for the quantative aptitude test.