Spam pages on the web use various techniques to artificially achieve high rankings in search engine results. Human experts can do a good job of identifying spam pages and pages whose information is of dubious quality, but it is practically infeasible to use human e#ort for a large number of pages. Similar to the Trust Rank algorithm [1], we propose a method of selecting a seed set of pages to be evaluated by a human. We then use the link structure of the web and the manually labeled seed set, to detect other spam pages. Our experiments on the WebGraph dataset [3] show that our approach is very e#ective at detecting spam pages from a small seed set and achieves higher precision of spam page detection than the Trust Rank algorithm, apart from detecting pages with higher pageranks [10, 11], on an average.