The problematic addressed in this article is dealing with the improvement of retrieval in Case-Based Reasoning for system design. The retrieval activity is based on the evaluation of similarities between requirements (target) and the solutions (sources). However, similarities between features is often a subjective kind of knowledge difficult to formalize within companies. Based on an ontology of domain, the approach permits to retrieve compatible solutions rather than similar ones using a model of designer preferences. The requirements are modeled by means of constraints. When constraints are confronted to solutions in order to evaluate a compatibility measure, missing information within solutions with regard to requirements are taken into account using semantic similarities between concepts. A case study validates the proposals.