Examination marking utilises a variety of cognitive processes, and from a psychological perspective, the demands that different questions place on markers will vary considerably. To what extent does marking accuracy vary among markers with differing backgrounds and experiences? More fundamentally, what makes some questions harder to mark accurately than others? The authors sought to address aspects of these key questions in an empirical study, which focused on GCSE mathematics and physics papers. For each subject, groups of "expert" and "graduate" markers were led by a Principal Examiner in the marking of identical samples of candidates' responses on a question-by-question basis. A quantitative analysis explored marking accuracy among questions and its potential relationships with marker type and question difficulty. Using their recently formulated model of cognitive marking strategies, the authors also investigated the relationship between accuracy and apparent cognitive marking strategy usage. Very few differences between experts and graduates were identified. However, variations in accuracy among individual questions (for all markers) were related to both question difficulty and apparent strategy usage: questions requiring markers to use more complex "reflective" thought processes were marked less accurately than those entailing only simple "intuitive" judgements. The findings have implications for the management of markers and for question design. It is concluded that, given adequate support and procedural training, graduate maths and physics markers may be able to mark almost all questions (of the kinds explored) as accurately as their expert counterparts can. (Contains 8 figures and 3 tables.)