In the past, academic interest inventory always use the classified groups to be discriminated as the multi-dimensional construct. However, due to the rapidly changing higher education environment, researchers do test developing faces the difficulty that the revision cycle of scale is shortened. This paper proposes another strategy. We hypothesize that using ”knowledge descriptor” instead of ”classified groups” is a better strategy. Databases of the Classification of Instructional Program (CIP) & O*NET are used to examine this proposition. Discriminant analysis shows that ”knowledge descriptor” has much higher discriminating power than Holland's RIASEC descriptor. It is also found that ”knowledge descriptor” has higher discriminating power than the ”skill descriptor” and the ”ability descriptor”. Implications of this revision strategy on academic interest inventory are further discussed.