We describe a study of the impact of exam question structure on the performance of first year Natural Sciences physics undergraduates from the University of Cambridge. The results show conclusively that a student's performance improves when questions are scaffolded compared with university style questions. In a group of 77 female students we observe that the average exam mark increases by 13.4% for scaffolded questions, which corresponds to a 4.9 standard deviation effect. The equivalent observation for 236 male students is 9% (5.5 standard deviations). We also observe a correlation between exam performance and A2-level marks for UK students, and that students who receive their school education overseas, in a mixed gender environment, or at an independent school are more likely to receive a first class mark in the exam. These results suggest a mis-match between the problem-solving skills and assessment procedures between school and first year university and will provide key input into the future teaching and assessment of first year undergraduate physics students.