This paper describes an evaluation of filtering methods for bilingual terminology extraction. Terminology extraction systems often favor recall over precision. This strategy results in an enormous number of term candidate pairs that have to be manually checked and cleaned. In the most extreme the post-editing step is so cumbersome that it prevents a system from practical employment. We show that filters based on formal criteria efficiently help in reducing manual labor. The most promising filter is based on the length difference between the source term candidate and the target term candidate.