In studies that employ matched pair analysis to identify environmental exposures important for a disorder, criteria for discordant pairs are seldom discussed. Yet several assumptions concerning the definition of discordancy may have considerable influence over what results are found. Problems are exacerbated when age of onset for a disorder is late in life. We propose a new set of criteria for defining discordant pairs in studies of dementia, taking into account duration of discordance and competing causes of mortality, and evaluate the consequences of choosing alternative definitions of discordancy.