Users seeking information in distributed environments of large numbers of disparate information resources are often burdened with the task of repeating their queries for each and every resource. Invariably, some of the searched resources are more productive (yield more useful documents) than others, and it would undoubtedly be useful to try these resources first. If the environment is federated and a single search tool is used to process the query against all the disparate resources, then a similar issue arises: Which information resources should be searched first, to guarantee that useful answers are streamed to users in a timely fashion. In this paper we propose a solution that incorporates techniques from text classification, machine learning and information retrieval. Given a set of pre-classified information resources and a keyword query, our system suggests a relevance ordering of the resources. The approach has been implemented in prototype form, and initial experimentation has given promising results.