AFQT scores than MCI-discordant pairs; MCI-discordant pairs had lower scores than normal-concordant pairs. Within-pair AFQT differences were observed only in dizygotic discordant pairs; normal twins scored higher. Overall results were similar for age 55 AFQT even after adjusting for age 20 scores. Conclusions: Neuropsychologically-defined MCI is present in the sixth decade of life (50s) in non-clinically referred, community-dwelling adults. A unique feature of the discordant twin design enabled us to identify unobserved heterogeneity. The subgroups of normal twins are phenotypically the same, but differed in premorbid cognitive ability based on genetic risk (i.e., having a normal or MCI co-twin). Differences based on genetic risk for the MCI phenotypes followed the same pattern. Premorbid ability is not the entire story, however, because differences persist even after adjusting for premorbid ability. Finally, genetic influences cannot cause differences in genetically-identical monozygotic twins. Therefore, the fact that premorbid cognitive differences within discordant pairs were accounted for by only dizygotic twins suggests that the link between premorbid cognitive ability and MCI is due to some degree of shared genetic influences.