The study objective was to evaluate the quality of parent-child interactions in preschool-aged children exposed prenatally to cocaine. African-American mothers and their full-term newborns (n = 343) were enrolled prospectively at birth and classified as either prenatally cocaine-exposed (n = 157) or non-cocaine-exposed (n = 186) on the basis of maternal self-report and bioassays. Follow-up evaluations at 3 years of age (mean age, 40 mo) included a videotaped dyadic play session and maternal interviews to assess ongoing drug use and maternal psychological distress. Play interactions were coded using a modified version of Egeland et al's Teaching Task coding scheme. Regression analyses indicated cocaine-associated deficits in mother-child interaction, even with statistical adjustment for multiple suspected influences on interaction dynamics. Mother-child interactions were most impaired in cocaine-exposed dyads when the mother continued to report cocaine use at the 3-year follow-up. Multivariate profile analysis of the Egeland interaction subscales indicated greater maternal intrusiveness and hostility, poorer quality of instruction, lower maternal confidence, and diminished child persistence in the cocaine-exposed dyads.