Interaction during feeding sessions between preterm infants and their mothers and that between full-term infants and their mothers were compared. (Mothers and babies were from a low-income, inner-city population.) When the children were about 3 years old, they attended a day camp for 3 weeks, during which their cognitive ability (Stanford-Binet) and social ability (both social competence and social participation) were assessed. Early interaction was quite different for preterms and full-terms, but in general it did not predict either social or cognitive ability at age 3. Birth status (preterm/full-term) did predict cognitive (but not social) ability: preterms scored lower. Finally, the children of mothers who were more emotionally and verbally responsive during a home visit at 20 months exhibited more social and cognitive ability at age 3. These results suggest that the baby, within broad normal limits, may be "buffered" against any long-term consequences of interaction during the first few months of life.