Air pollution is an important determinant of health and poses a significant threat globally. The World Health Organisation (WHO) are at the forefront of health modelling and policy development worldwide and must ensure that this is based on accurate and convincing evidence. A coherent framework for integrating data from various sources is required that provides accurate and effective analysis and yield exposure estimates with associated uncertainty. These estimates should be consistent with raw data and provide a means for explanation when there are discrepancies. I will explore the current methodology used within WHO to estimate air pollution levels and how changes in framework can significantly improve model predictions. I will also explain how Bayesian melding can be used to match the requirements for air pollution modelling within WHO and the associated challenges that arise from using this technique.