We examined the association of proximal and distal training gain to subsequent mental status ratings in 302 participants (M = 76.62 years) trained on inductive reasoning or spatial orientation in the Seattle Longitudinal Study. Only training effects on reasoning ability were predictive of mental status group membership. Participants subsequently rated as probably demented did not significantly differ from nondemented participants in magnitude of reasoning training gain 14 years prior to assessment, but they did 7 years prior to status ratings. Proximal training gain 1 year prior to assessment was 0.40 SD for nondemented participants, compared with 0.25 and 0.10 SD for at-risk and probably demented participants, respectively. The combination of reasoning ability training and increased proximal training gain on reasoning ability was associated with a decreased likelihood of being rated as probably demented.