The number of available dss within organizational intranets will soon require efficient retrieval functionality. While web retrieval technology performs excellent on documents, computational services need approaches that capture the semantics of resources. We present a retrieval approach that uses a variant of structured modeling to represent resources. It allows the use of similarity of models for retrieval. Exact similarity computation is shown to be np-hard, and efficient heuristics for similarity computation and filter algorithms are introduced. We report an evaluation in a classroom experiment and give computational results on a benchmark library.