Across England and Wales, when a person is booked into police custody they are assessed for risk of harm to themselves or to others. This is an important process as detainees often present with complex physical, mental and emotional needs and each year a number die or self-harm whilst in custody. The purpose of this initial risk assessment is to mitigate that risk through risk identification and treatment, in the form of observation levels. National guidance provides a set of recommended questions to ask the detainee, and also four observation levels ranging from hourly checks to constant observation. However, evidence has shown that there is variation in the questions asked, both away from national guidance and between police forces, as well as variation in observation levels set. To remove the subjectivity in decision making within this risk assessment process, with a move towards a decision support tool for treatment determination, understanding of the risk drivers is initially required. To the authors’ knowledge, there has been little research into understanding the current decision mechanisms and also the level of consistency between risk identification and treatment. Therefore the purpose of this study is to use data from custody records to understand which risk factors have the greatest impact on the observation levels that custody officers set. To identify which are the key factors that appear to have the most impact on determining observation level, modelling methods such as regression, discriminate analysis and classification trees have been used to develop statistical models which have then been compared in terms of ability to predict observation levels based on these factors. The findings from this study will have a practical impact on understanding the risk assessment process within police custody and facilitate a more robust process.