This paper reports on an evaluation of an avatar for Swiss German Sign Language (Deutschschweizerische Gebärdensprache, DSGS) by native signers of this language. The avatar is the final output of a machine translation system which is being built to translate in real time German announcements of the Swiss Federal Railways (Schweizerische Bundesbahnen, SBB) into DSGS. An existing software, JASigning, is used to automatically generate the avatar animations. The evaluation data is from a focus group with seven Deaf signers who provided feedback on how to improve the DSGS avatar. They identified several aspects that had to be modified: Among them were the color of the avatar's clothing and the background, the direction of the avatar's default eyegaze, the speed of fingerspelling and mouthings, the temporal coordination of the manual and non-manual components of a sign, and the handling of lists of signs. These aspects are not just relevant for train announcements in DSGS, but for data of other kinds and other sign languages as well.