本研究的目的是:應用概化理論估計多層面變異來源的運動技能測驗。五十九位受試者實施壘球打擊成績表現測驗。採受試者、試做次數和評分者等三個層面的重複設計，以電腦程式GENOVA進行估計。概化研究的結果是:一、測驗的真實變異佔總變異成份的22.15%。二、「受試者和試做」交互作用的變異成份佔總變異的70.55%'烏最主要的變異來源三、其餘的變異來源所佔的比率很小。決斷研究的結果發現:增加試做次數可以有效的提高概化係數，而增加評分者人數對測驗的概化係數增加不多。根據研究的結果提出兩點結論:l.多層面變異來源的運動技能測驗應用概化理論，可以得到整體測驗的概化信度係數。2.概化研究可以了解各種變異來源佔總變異成份的比率，依測驗使用者的需求，決定最適當的測驗條件配置。

The purpose of this study was to apply generalizability (G) theory deciding the criterion for trials-to-criterion tests in the psychomotor test. The subjects were three trials-to-criterion tests. The object of measurement is subjects variance. The designs of this study are a one-facet crossed-design. We approach different sources of score variation (subjecs， criterion and interaction) and the phi coefficient of multipIe criterion by using a GENeral purpose analysis Of Y Ariance system (GENOY A). The conclusion were: 1.The true variance of tests are acceptable and the percentage of total variance are from 12.5 to 16.1 %. 2. The criterion variance of tests are small. Theyare from 0.0 to 2.2% in the percentage of total variance. Therefore， it is very similar between G coefficient and phi coefficient. 3. The main soure of variation is residual error of test. they are from 81.7 to 87.5 %. 4. The trials-to-criterion tests of psychomotor test are: (1)From，c6 to 7 attempts of criterion yield mediocre levels of estimated phi coefficient. (2) From 21 to 28 attempts of criterion yield reasonable levels of estimated phi coefficient.