Humphreys fails to consider the complex relations between genotypes and phenotypes for intelligence. I use the term genotype in this context in both commonly accepted definitions-as a latent characteristic of a person and as the biological genotype determined for an individual at the moment of conception. If genotypes for intelligence exist that are independent of the level of intelligence expressed in the phenotype of an individual, it may be necessary to consider both the phenotypic and the genotypic level of intelligence of a person in order to understand changes in intelligence and the design of appropriate interventions to change the impact of the genotype on the phenotype. There are both conceptual and empirical reasons to distinguish between phenotypic and genotypic levels of intelligence. There are manifestations of genotypes present in the first year of life that are predictively related to measures of childhood intelligence obtained as late as 8 years. Measures of habituation and response recovery obtained in the first 6 months of life correlate with childhood IQ scores in excess of .4, uncorrected for attenuation (Bornstein, 1989). These data imply that the development of phenotypic levels of intelligence is predictable from a measure of information processing before the development of a sufficiently large repertoire of phenotypic intelligence that would permit one to derive an appropriate index of intelligence by sampling from the repertoire. These data may not be compatible with the assumption that the correlation for tests over occasions forms a quasi-simplex matrix. Studies of twins reared together and apart and adoption studies provide evidence for a changing relation between phenotypes and genotypes for intelligence over the life span. The heritability of intelligence is a monotonically increasing function of age. Correlations of IQ scores for biologically unrelated siblings reared together and between the IQs of adoptive parents and the IQs of their adopted children decline to near-zero values as the adopted children grow older (Brody, 1992, chap. 5; Loehlin, Horn, & Willerman, 1989). Correlations between the IQs of biological parents and their adopted children exhibit little or no decline over time. Behavior genetic analyses of these data imply that the heritability of IQ increases with age and that the influence of the shared environment declines. IQ correlations for monozygotic twins increase in childhood and appear to remain constant over the life span. IQ correlations for dizygotic twins, by contrast, decline over the life span (Brody, 1992; McGue, Bouchard, Iacono, & Lykken, in press). If the heritability of IQ is a monotonically increasing function of age, then changes in IQ over the life span may be construed as changes in phenotypes that increase the similarity between a phenotypic score and a biological genotype present for each person at the moment of conception. The analysis of the relation between changes in phenotypic intelligence and genotypic dispositions may have implications for the design of intervention programs. Modifications in the environment provided early in life may be of diminishing importance as the effects of the early environment fade and as genotypic dispositions increasingly determine the growth and development of intelligence. We know that early interventions have diminishing influences on intelligence over time (Brody, 1992, chap. 6; Consortium for Longitudinal Studies, 1983). Comprehensive interventions that start shortly after birth and continue for the first 5 years of life lead to changes in intelligence that fade over time. The Abecedarian Project, for example, has reported gains in intelligence of approximately one third of a standard deviation at age 12 (Ramey, 1993). Even adoption effects for older adoptees are weakwith estimated effect sizes that may be between .00 and .5 SD. Turkheimer (1991) analyzed French adoption studies and derived a regression equation for the relation between the educational background of the adop-