We study coercion-resistance for online exams. We propose two properties, Anonymous Submission and Single-Blindness which, if hold, preserve the anonymity of the links between tests, test takers, and examiners even when the parties coerce one another into revealing secrets. The properties are relevant: not even Remark!, a secure exam protocol that satisfied anonymous marking and anonymous examiners results to be coercion resistant. Then, we propose a coercion-resistance protocol which satisfies, in addition to known anonymity properties, the two novel properties we have introduced. We prove our claims formally in ProVerif. The paper has also another contribution: it describes an attack (and a fix) to an exponentiation mixnet that Remark! uses to ensure unlinkability. We use the secure version of the mixnet in our new protocol.