Cross language information retrieval (CLIR) is to find all relevant documents for a given query in a situation where the query language is different from the document language. Because the query and document languages differ, a CLIR system necessarily includes a translation step. The results of several intermediate retrieval runs must then be combined into one result list but their absolute scores cannot be compared directly. The discussion of these problems and how they were dealt with will be illustrated by the CLIR results of the Twenty-One group (TNO-TPD and Twente University) at TREC6, TREC7 and TREC8.