"Item difficulty modelling" is a technique used for a number of purposes such as to support future item development, to explore validity in relation to the constructs that influence difficulty and to predict the difficulty of items. This research attempted to explore the factors influencing question difficulty in a general qualification using item difficulty modelling and to explore the usefulness of the method for examinations in this context. The study modelled the difficulty of the questions in a multiple-choice A level physics examination. A range of variables relating to the questions and the processes judged to be involved in answering were identified and the questions were coded for these variables. These were then used in statistical models (multiple linear regression and the Rasch Linear Logistic Test Model (LLTM)) to attempt to explain the variance in difficulty of the questions. The regression analysis was able to explain 89% of the variance in difficulty estimates for the questions modelled and 66% when adjusted for the relative number of questions and question features. The question features identified as significant predictors of question difficulty were the total amount of reading, and requirements for the student to use physics concepts, work with symbols and carry out calculations. The LLTM was able to predict 81% of the variance in difficulty estimates for the questions modelled, but only 41% of variance once adjusted for the number of question features and questions. The LLTM found many of the question feature variables to be significant predictors due to its greater statistical power, but the effects of many of the individual variables were unstable. This has implications for the usefulness of item difficulty modelling with assessments which do not involve item banking and reuse of items such that many items can be modelled together. (Contains 4 tables and 2 figures.)