A total of 403 healthy, premenopausal women, residing near Santa Clara, California, were recruited from a large health care plan in California for a study of menstrual function. After a telephone interview, participants collected daily urine samples and recorded bleeding and other information in diaries. Data were collected during 1990-1991. Urine samples were analyzed for creatinine and for estradiol and progesterone metabolites by enzyme-linked immunoassay. Computer algorithms were developed to derive menstrual segment length, ovulatory status, day of ovulation, and other parameters from the urine and diary data. (We use "segment" rather than "cycle" to avoid implying that normal cycling occurred.) The average length of participation was 141 (standard deviation, 45) days. The mean segment length was 28.8 (standard deviation, 4.4) days; follicular phase length, 16.0 (standard deviation, 4.4) days; and luteal phase length, 12.9 (standard deviation, 1.7) days; 19 (4.7%) women experienced anovulatory episodes. In exploratory multivariate analyses, important associations included the following: age of > or = 35 years with decreased segment and follicular phase lengths; heavier weight (upper quartile) with anovulation and increased follicular phase and decreased luteal phase lengths; Hispanic ethnicity with anovulation and increased segment length; and past difficulty in achieving pregnancy with anovulation and increased length and variability of segments and follicular phases. Urine biomarkers can be used successfully to evaluate menstrual function in epidemiologic studies.