We focus on the problem of filtering a fragment of the knowledge contained in a large conceptual schema. The problem appears in many information systems development activities in which people need to operate with a piece of the knowledge contained in that schema. We propose a new method in which a user focuses on one or more entity types of interest for her task at hand, and the method automatically filters the schema in order to obtain a set of entity and relationship types (and other knowledge) relevant to that task, taking into account the interest of each entity type with respect to the focus, computed from the measures of importance and closeness of entity types. The method has been implemented in a prototype tool, and it has been experimented with the schema of the osCommerce and the ResearchCyc ontology.