The combination of evidence for Information Retrieval has been studied extensively in order to increase effectiveness. In this paper, we study the selective application of different retrieval approaches on a per-query basis for Web Information Retrieval. Our methodology is based on the assumption that not all queries benefit from a uniform retrieval approach. In order to select the most appropriate retrieval approaches for each query, we use evidence from the hyperlink structure of the retrieved documents and also from the distribution of aggregates, that is the groups of documents from the same domain. Our experimental results show that it is possible to obtain important improvements in retrieval effectiveness by using simple statistical decision mechanisms on a per-query basis.