Link analysis has been used to enhance retrieval results of the web search for years. PageRank and HITS are the two well-known algorithms widely used by most researchers. The former analyzes the web links off-line, but does not consider either web topics or user's query. The latter judges the web page on-line according to the user's query. Here we propose a novel algorithm, called "Topic-Centric", which also performs link analysis of web pages off-line. The authoritativeness of web pages is calculated in accordance with the web topics. To test the effectiveness, we apply the proposed algorithm to re-rank the search results given by a traditional retrieval system. Preliminary experiments using standard web collections, though inconclusive, show interesting retrieval results.