Since the early in 1990's when frequent commercial use of the Internet starts, research on websites has been actively conducted both by academic researcher and by website practitioners. Three broad categories of the research are Web content mining, Web structure mining, and Web usage mining. Apart from those, there are some research on coloring, placement technique of images, texts, and links on each page. In this paper, we focus on the difference between two cluster structures of website, one induced from link-based property and the other from term-based property. The link-based property is stable until a new link is added, but the term-based one varies depending on the items for searching. We propose an evaluation method for website by comparing the structures of clusters resulted from these properties. As the clustering method, here we adopt kernel k-means method, and compare partial clusters derived from term-based property depending on the given sequence of particular terms to de inite partial clusters from link-based property. In order to distinguish them, we try to adopt spectral analysis.