A study was carried out to compare the performance of enzyme-linked immunosorbent assay (ELISA) and blood polymerase chain reaction (PCR) for diagnosis of paratuberculosis in cattle and sheep. For cattle, a set of 278 samples from 1 paratuberculosis-affected Friesian farm was used; it included 80 ELISA-positive samples and 198 ELISA-negative samples from an age-matched group. Ninety-four samples were from heifers and 184 were from 2-5-year-old cows. The overall analysis showed a clear association (Fisher exact test [FET] P = 0.0049) but a weak negative agreement (45.3%, kappa = -0.1665 +/- 0.0994) between the 2 tests. It reflected a moderate agreement among heifers (87.7%, kappa = 0.4471 +/- 0.2435) and a moderate disagreement among cows (62.7%, kappa = -0.3670 +/- 0.1057). For sheep, 496 blood samples from 53 Latxa dairy flocks were used; 180 of the blood samples were from dam/offspring pairs. The overall association between the 2 tests on ovine samples was strong (FET, P = 0.0005), whereas the agreement was low (kappa = 0.1622 +/- 0.1188). There was slightly better agreement for ewes (kappa = 0.2135 +/- 0.1992) than for lambs (kappa = 0.1193 +/- 0.1301). There was also a highly unlikely proportion of dam/offspring positive results (FET, P < 0.0001, kappa = 0.6269 +/- 0.1854). Four of 6 lambs that were necropsied 1 year after testing had paratuberculosis microscopic lesions in the ileocecal valve (3 lambs) or a PCR-positive result (4 lambs). These results suggest that blood PCR testing might be a potentially useful new approach in paratuberculosis diagnosis, especially in young animals.