Twin analyses of phenotypes that are associated with mortality may provide biased heritability estimates if the models require that data from both members of a pair are available. This is particularly true when longitudinal analyses are applied to measures of cognition or biomarkers of aging. The effect of applying imputational techniques that include information on age at death was tested on longitudinal data from two twin studies of aging, each with up to four occasions of measurement. Measures of twin similarity for intercepts and slopes from three latent growth curve models were compared: without imputed data, including imputed data but without information on age at death, and including imputed data with information on age at death. Results indicated that twin similarity for slopes decreases when mortality is accounted for, but that considerable age-related covariation remains.