Currently we prepare examination papers using Microsoft Word. It is well known that word processors are low productivity compared with semantic markup. It is also clear that because a word processor is ready to let you write anything, it has very limited ability to detect structural or semantic errors (such as leaving out question numbers or marks). But there is worse: there are many people who have a legitimate interest in reading an examination paper, but a word processor enforces "What Anyone Sees Is What Everyone Gets". I will explain the SGML "a-document-is-a-database" idea and the EXAM Document Type Description I have been working on, and argue that we can and should switch to SGML for producing exam papers.