A field study was conducted in 10 commercial dairy herds to test the efficacy of a rapid enzyme-linked immunosorbent assay to identify errors in detection of estrus and diagnosis of pregnancy. Milk samples were collected at the milking following detection of estrus and on d 21 postinsemination. Progesterone was measured on-farm by an enzyme-linked immunosorbent assay and in the laboratory by radioimmunoassay. Agreement between the two methods of progesterone analysis (n = 820) was 89.8%. There were 93 cows (18.7%) by laboratory analysis and 126 cows (25.4%) by on-farm analysis for which concentration of progesterone in the sample collected at suspected estrus (n = 497) was higher than normal estrual values. Exclusion of these cases increased the overall accuracy of pregnancy diagnosis by 6.8 and 6.4% for laboratory and on-farm methods. Overall agreement between pregnancy diagnosis by milk progesterone at d 21 postinsemination and diagnosis by return to estrus or palpation of reproductive organs was 80.5 and 75.8% for laboratory and on-farm methods (75.0% and 69.7% for pregnant cows and 94.5 and 93.9% for nonpregnant cows). These results indicate that an enzyme-linked immunosorbent assay for milk progesterone performed on-farm was equal in accuracy to a laboratory radioimmunoassay for identifying errors in detection of estrus and diagnosis of pregnancy.