We use a statistical model that combines longitudinal and survival analyses to estimate the influence of level and change in cognition on age at death in old and very old individuals. Data are from the Berlin Aging Study, in which an initial sample of 516 elderly individuals with an age range of 70 to 103 sears was assessed up to 11 times across a period of up to 13 years. Four cognitive ability domains were assessed by two variables each: perceptual speed (Digit Letter and Identical Pictures), episodic memory (Paired Associates and Memory for Text), fluency (Categories and Word Beginnings), and verbal knowledge (Vocabulary and Spot-a-Word). Longitudinal models on cognition controlled for dementia diagnosis and restest effects, while survival models on age at death controlled for age, sex, socioeconomic status, sensory and motor performance, and broad personality characteristics. Results indicate: (1) Individual differences in the level of and in the linear change in performance are present for all cognitive variables; (2) when analyzed independently of cognitive performance, all covariates, except broad personality factors, predict survival; (3) when cognitive performance is accounted for, age, sex, and motor performance do predict survival, while socioeconomic status and broad personality factors do not, and sensory performance does only at times; (4) when cognitive variables are analyzed independently of each other, both level and change in speed and fluency, as well as level in memory and knowledge predict survival; (5) when all cognitive variables are analyzed simultaneously using a two-stage procedure, none of them is significantly associated to survival. In agreement with others, our findings suggest that survival is related to cognitive development in old and very old age in a relatively global, rather that ability-specific, manner.