An approach to testing reading comprehension among Dutch speaking university students of German for academic purposes is described. The test items were designed to reflect the most typical comprehension errors made by speakers of Dutch when reading German academic discourse. The most frequent errors were due to vocabulary and declension difficulties. After reading a German language text, students are given open-ended questions about excerpts involving these problem areas. The questions require the students to provide loose translations of phrases from the text. Use of the first language for response eliminates the effect of productive skills. The open-ended items discriminate better than multiple choice items because the context is necessary for the right solution. The approach and variations have been tried several times with a group of 23 students, and the results compared and analyzed for reliability and validity. Reliability was found to be high, and correlations between detailed and global comprehension were low. Choice of test and item content is discussed. (RW)