Abstract In proposing a searching strategy in a hypertext environment, we have considered three criteria: (1) the retrieval process should use multiple sources of evidence, including multiple indexing schemes and multiple search mechanisms; (2) the hypertext links should be exploited in order to find more relevant material and "good" starting points for browsing; and (3) the user's information need must be easily expressed. To satisfy these three criteria, we have implemented a hypertext system where the user, instead of writing a formal query, just selects relevant terms and/or documents. Based on multiple indexing schemes, Bayesian networks are used to manage the indexing spaces and to store the user's information need. To combine multiple search techniques, the hypergraph is also composed of implicit links (bibliographic coupling, co-citation, etc.), and computed links storing the nearest neighbors of each node. Using link semantics, a constrained spreading activation retrieves relevant nodes for browsing.