Although caffeine is commonly consumed during pregnancy, there are few reports on the association of in utero caffeine exposure with offspring cognition or behavior during childhood. We evaluated the association of maternal serum paraxanthine, caffeine’s primary metabolite, at <20 and ≥26 weeks’ gestation with the child’s intelligence quotient (IQ) and problem behaviors at ages 4 and 7 years among 2,197 mother-child pairs. The mothers were controls from a case-control study of caffeine metabolites and spontaneous abortion that was nested within the Collaborative Perinatal Project (multiple US sites, 1959–1974). Associations of paraxanthine (adjusted for maternal age, race, education, smoking, prepregnancy weight, gestational age at blood draw, and child sex) with mean IQ were assessed by linear regression and associations with problem behaviors by logistic regression. Paraxanthine concentration at ≥26 weeks’ gestation manifested an inverted-J-shaped association with child’s IQ at age 7 years, with a peak difference (vs. undetectable) of 0.65 points at 750 µg/L (66th percentile) and a decrement thereafter. Paraxanthine at <20 weeks was linearly associated with internalizing behavior at age 4 years (for a 500-µg/L increase, odds ratio = 1.3, 95% confidence interval: 1.1, 1.5). None of the remaining 12 associations approached statistical significance. We conclude that over a range of values applicable to most pregnant women, there was no meaningful association of serum paraxanthine level with childhood IQ or problem behaviors.