Few empirical data exist on the characteristics of subjects who provide semen specimens in epidemiologic studies. The objective of this investigation was to determine participation rates and potential biases in a contemporary study of human semen quality. Subjects (n = 268) are a subset of the Child Health and Development Studies. Their mothers enrolled between 1960 and 1963 during pregnancy. Archived prenatal serum samples, prenatal and birth records, placental examinations, and follow-up for growth and development through adulthood are available. Sons were aged 36-39 years at the time of this study. Respondents to the initial mailing and nonrespondents, who were subsequently traced and recruited, differed in semen parameters, including sperm concentration (78.3 x 10(6)/ml for respondents vs. 37.2; p = 0.003) and the percentage of normal morphology according to the 1987 criteria of the World Health Organization (58.7% for respondents vs. 53.3%; p = 0.04). The authors conclude that researchers designing population-based studies of semen parameters should expect nonrepresentative samples. Adaptation of the design to anticipate and mitigate bias and to maximize efficiency can yield scientifically sound information. Recommendations for study designs are discussed.