本研究透過高階層驗證性因素分析與模式競爭，為標準化空間能力測驗找尋最 佳的最簡約模式，因為根據該測驗的編製發展歷程，共有七個可能的理論模式，從 一階、二階到三階均有。本研究樣本資料為臺灣地區高一學生1,641 人，測驗共計 18題，分屬四個題型與兩個空間能力因素，測驗資料屬二元計分變項。 本研究同時採用最大概似估計法和對角線加權最小平方法進行參數估計，以比較 不同參數估計方法對於分析結果的影響。結果發現，對角線加權最小平方法能估計與 適配的模式較多，共有六個模式。而在最佳模式的競爭上，兩種估計法皆選擇同樣的 一個二階模式架構，其第一階潛在變項為四個題型，第二階潛在變項為空間能力。 研究結果確認該測驗具高階層模式架構，但是其他的高階模式不盡然比低階模 式更為簡約，未來在測驗的應用上，本研究結果的二階層模式能有最好的解釋。惟 研究結果可能受限於該測驗在因素、題型與試題數量上的限制，未來修訂測驗時， 若能適度擴增因素、題型與試題數量的組合，其高階層結構的穩定性將更為可期。

The purpose of the study is to find the best parsimonious explanatory model for the Standardized Spatial Ability Test (SPAT) through higher-order confirmatory factor analysis and model competition. This is because there are seven possible structure models including first-, second-, and third-order models according to the test development and construction of SPAT. The data used is the norm of SPAT, which is a sample of 1,641 first-year senior high school students. There are 18 binary test items categorized into four item types within two spatial factors. For parameter estimation, maximum likelihood estimation (ML) and diagonally weighted least squares (DWLS) are both used to compare their differential impact. Results show that there are six models fitted well by DWLS, more than those fitted by ML. Through model competition, the best parsimonious model chosen by both ML and DWLS is the same second-order structure of which, the first-order latent variable is the four item types, and the second-order is the spatial ability. It is confirmed in the study that higher-order structure models exist in SPAT, however other higher-order structure models are not necessarily better than the lower-order ones. The chosen second-order structure can provide best explanation in future SPAT applications. The results may be restricted by the limited numbers of spatial factors, item types, and items in SPAT. If there is revision of SPAT in the future, increasing the spatial factors, item types and items would lead to more promising and stable higher-order structure.