Within and between subject variability in ASL perfusion data can obscure subtle perfusion changes indicative of neurodegenerative disease. It prohibits the production of a 'healthy' population-average perfusion atlas to which individuals can be compared. This work used a GLM to produce subject-specific perfusion references based on tissue partial volumes. Results show that such a linear model can successfully decompose perfusion, at the population level, into components related to tissue partial volumes. However, the model struggles to capture variability at the individual subject level and thus, the problem will in future be addressed with non-linear modelling techniques.