We applied data mining techniques to explore survival in a sample of 6'203 adults (age range 42-93 years), living in the Manchester and Newcastle-upon-Tyne (UK.) areas. We were particularly interested in the relations between cognitive performance and mortality prediction. Participants were assessed up to four times over 20 years on several psychological and health-related variables and were also administered an extensive battery of cognitive tasks. We applied linear mixed models to estimate level of cognitive decline and change (mostly decline) therein for each individual. We then utilized Cox proportional-hazards modeling to predict time to death based on levels of and changes in cognitive performance, and on demographic and social predictors. Next, to gain further insight into the survival process, we used recently developed induction trees and ensemble methods. These models allow studying complex and asymmetric interactions and non-additive functions of model predictors. Particularly relevant to our theoretical purposes, the random forest approach allowed us to identify a set of demographic and cognitive variables that strongly in uenced survival. We conclude that induction trees and ensemble methods are a useful extension to more classical models in that they are not limited by common modeling assumptions and can reveal complex patterns of relation.