The purpose of this study was to explore the relationship between the density of LBs in the brain and cognitive deficits. This question was investigated by analysis of correlations between the densities of LBs in the entire study cohort and in subgroups of cases with and without AD-type neuropathological disorders. It should be emphasized at the outset that the great majority (83%) of the 273 autopsy cases studied did not have LB inclusions in any of the brain regions studied. Forty-six (17%) of the 273 subjects did have LB inclusions in one or more brain regions. The density of LBs was positively correlated with the degree of cognitive deficits as measured by the CDR without regard to the presence or absence of other neuropathological lesions. The relationship between cognitive deficits and LB density was initially studied by deriving a number of summary variables. These variables included the sum of CERAD ratings (LB-CERAD) (0, none; 1, sparse; 3, moderate; and 5, severe) of 273 subjects with LB densities in the brain regions studied, and the sum of the quantitative direct counts of LBs (LB-Cortex) in the 5 neocortical regions outlined above in a randomly selected subgroup (n = 145). These 2 summary variables were significantly and positively correlated with each other (LB-CERAD vs LB-Cortex: r = 0.76, P<.001, n = 145), and each was significantly and positively correlated with the CDR score (LB-CERAD vs CDR: r = 0.19, P = .002, n = 273; LB-Cortex vs CDR: r = 0.31, P<.001, n = 145). The correlations between LB density and CDR were stronger when only those cases with LBs were considered (LB-CERAD vs CDR: r = 0.34, P = .02, n = 46; LB-Cortex vs CDR: r = 0.58, P<.001, n = 34; Figure 1). The correlation of LB-CERAD with CDR remained significant (r = 0.16, P = .002, n = 208) when the subjects with severe dementia (CDR, 5) were eliminated from the analysis. The correlation of LB counts in the cortex alone did not reach statistical significance (r = 0.17, P = .07, n = 11) when the data from subjects with severe dementia (CDR, 5) were excluded from the analysis. Analysis of variance confirmed the significant increase in cortical total LB counts as a function of CDR (F6,138 = 3.53; P = .003). The Duncan multiple range test showed that the density of LBs in subjects with a CDR score of 5 was significantly higher (P<.02) than the density of LBs in the cortices of all other CDR groups, except for subjects with a CDR score of 4, who did not differ significantly (P<.08) from the cohort with a CDR score of 5. Because the LB-Cortex count-sum variable may not have been normally distributed, these analyses were repeated by performing Kruskal-Wallis, Mann-Whitney, and Spearman rank correlation nonparametric procedures. The statistically significant relationships described above were replicated with these nonparametric tests.