The goal of zero-knowledge watermark detection is to allow a prover to soundly convince a verifier of the presence of a watermark in certain stego-data without revealing any information which the verifier can use to remove the watermark. Existing proposals do not achieve this goal in terms of definition (not formally zero-knowledge), security (unproven) and coverage (handle only blind watermark detection). In this paper we define zero-knowledge watermark detection precisely. We then propose efficient and provably secure zero-knowledge protocols for blind and non-blind versions of a well-known class of watermarking schemes. Using these protocols the security and efficiency of many watermark based applications can be significantly improved. As an example of use we propose concrete protocols for direct proof of ownership which enable offline ownership proofs, i.e., copyright holders can prove their rightful ownership to anyone without involving a trusted third party in the actual pro...