Brains of non-demented elderly people were divided into two groups according to the presence or absence of senile plaques (SP(+) and SP(-)). The regional number of NFT in each group were then quantitatively investigated and compared with that in the late-onset Alzheimer's disease (AD) group and the limbic NFT dementia (LNTD) group. NFT were divided into type 1, type 2 and type 3 according to the developmental stage. In addition, polymorphism of the apolipoprotein E (Apo E) gene was analyzed in all groups. The most frequent regions of NFT common to all groups were the transentorhinal cortex, the entorhinal cortex, the subiculum and cornu ammonis (CA)1 of the hippocampus. In the SP(+) group the proportion of type 3 was high in the transentorhinal cortex and entorhinal cortex, while type 1 or 2 were high in the subiculum and CA1, suggesting that NFT formation progresses from the parahippocampal cortex to the hippocampus. In the SP(-) group the proportion of type 3 was higher in the subiculum and CA1 than in the transentorhinal cortex and entorhinal cortex, suggesting that NFT formation is accelerated in the hippocampus. The late-onset AD group and LNTD group showed the patterns of NFT formation similar to those of the SP(+) group and SP(-) group, respectively. The frequency of the epsilon4 allele of the Apo E gene was significantly higher in the late-onset AD group and SP(+) group than in the LNTD group and SP(-) group, respectively. From these findings it is suggested that persons in the SP(+) group are likely to remain non-demented elderly persons or become a developmental matrix of late-onset AD with a risk factor of the epsilon4 allele, while those in the SP(-) group are likely to remain non-demented elderly persons or pass into LNTD without a risk factor of the epsilon4 allele.