The Trivers–Willard hypothesis predicts that females in good condition should bear more sons rather than daughters in certain mammals, including humans. This study tests the hypothesis by using 66,638 childbirth records from a national birth cohort survey in current Japan. Our analyses showed that, contrary to the hypothesis, indicators of parental condition, such as mother’s age, body mass index, job status, education level, medical history, or household income, had few statistically significant effects on infant sex at birth. In previous studies investigating the Trivers–Willard hypothesis, the results have been quite mixed and inconclusive. We discuss some theoretical and methodological challenges towards a precise understanding of the hypothesis in human populations.