4 successive cohorts of low-income families were randomly assigned either to a home-based intervention program that focused on modeling verbal interaction between mother and child around selected toys and books or to comparison treatments. Large program effects were found on maternal interaction styles in videotaped observations. Small IQ and program-specific effects were found for children in contrast to much larger IQ effects found in earlier research. IQ effects did not appear to have been mediated by changes in maternal behavior. A variation in which toys and books were supplied without home visits was as effective as the full program on IQ but not on maternal behavior. 3 years postprogram , there were no detectable effects in achievement or IQ tests or in first grade teachers' ratings of school adjustment and performance, but IQ and achievement scores were near national norms. Reasons for discrepancies with earlier results are discussed. The results highlight the need for continued experimental evaluation of early intervention programs with safeguards to insure that samples are educationally at risk.