本研究的目的是探討負Alpha信度係數問題，並尋求適當的表示運動技能測驗的信 度。 從測量理論分析 Alpha 信度係數的背景，再以運動技能重複測量單一項目測驗的實例 說明，比較古典測量理論的 Alpha 係數、組內相關係數和 Rasch 個別信度係數的誤差，結 果發現：在受試者同質性高且重複測量單一項目（ MASI ）的運動技能測驗情況下， Alpha 係數容易得到負值， 而 Rasch 個別信度係數能有效的估計測驗信度。 本研究的結論為： Rasch 個別信度係數（ separation reliability coefficients ）能夠有效的克服負 Alpha 係數的問題。

　　　　　The purpose of the study was to investigate the meaning of negative coefficient alpha as an index of psychomotor tests. The background of psychometric theories was analyzed to the coefficient alpha for reliability. An illustration of the test of multiple-attempt, single-item had introduced. The errors of test scores were compared to the coefficient alpha and intraclass correlation coefficient of classical test theory and Rasch separation reliability coefficients. The result indicated that negative value of generating in the methods of coefficient alpha. They are not satisfaction both the high homogeneity of subjects and Multiple-Attempt, Single-Item Tests in the psychomotor. But, Rasch separation reliability coefficients can be valid estimation a tests' reliability. The conclusion of this study was that the Rasch separation reliability coefficients were valid overcome with the problem of negative coefficient alpha.