The Million Deaths Study (MDS) is a large health survey where detailed data on every death in 8000 sampling units (portions of villages or towns) are collected annually. The population in each year, by 5-year age and sex group, is known, as is risk factor information on a family member of each deceased individual. The talk will develop two Geostatistical models for quantifying the relationship between air quality and mortality, using each of the population denominators and suitably chosen control group.  Different analyses lead to sometimes conflicting results, and a detailed evaluation of the different modelling assumptions will be presented.