Background: Preliminary evidence suggests that prenatal testosterone exposure may be associated with language delay. However, no study has examined a large sample of children at multiple time-points. Methods: Umbilical cord blood samples were obtained at 861 births and analysed for bioavailable testosterone (BioT) concentrations. When participating offspring were 1, 2 and 3 years of age, parents of 767 children (males = 395; females = 372) completed the Infant Monitoring Questionnaire (IMQ), which measures Communication, Gross Motor, Fine Motor, Adaptive and Personal-Social development. Cut-off scores are available for each scale at each age to identify children with "clinically significant" developmental delays. Chi-square analyses and generalized estimating equations examined longitudinal associations between sex-specific quartiles of BioT concentrations and the rate of developmental delay. Results: Significantly more males than females had language delay (Communication scale) at age 1, 2 and 3 years (p-values less than or equal to 0.01). Males were also more likely to be classified as delayed on the Fine-Motor (p = 0.04) and Personal-Social (p less than 0.01) scales at age 3 years. Chi-square analyses found a significant difference between BioT quartiles in the rate of language delay (but not Fine-Motor and Personal-Social delay) for males (age 3) and females (age 1 and 3). Generalized estimating equations, incorporating a range of sociodemographic and obstetric variables, found that males in the highest BioT quartile were at increased risk for a clinically significant language delay during the first 3 years of life, with an odds ratio (OR) of 2.47 (95% CI: 1.12, 5.47). By contrast, increasing levels of BioT reduced the risk of language delay among females (Quartile 2: OR = 0.23, 95% CI: 0.09, 0.59; Quartile 4: 0.46, 95% CI: 0.21, 0.99). Conclusion: These data suggest that high prenatal testosterone levels are a risk factor for language delay in males, but may be a protective factor for females.