The topics of trust and privacy are morerelevant to users of online communities than ever before. Trust models provide excellent means for supporting users in their decision making process. However, those models require an exchange of informationbetween users, which can pose a threat to the users' privacy. In this paper, we present a novel approach fora privacy preserving computation of trust. Besides preserving the privacy of the recommenders by exchanging and aggregating recommendations under encryption, the proposed approach is the first that enables the trusting entities to learn about the trustworthiness oftheir recommenders at the same time. This is achievedby linking the minimum amount of information thatis required for the learning process to the actual recommendation and by using zero-knowledge proofs forassuring the correctness of this additional information.