An indirect ELISA was evaluated for the detection of Brucella antibodies in milk (m-ELISA) from sheep experimentally infected with B. melitensis biovar 3. At the end of the second reproductive cycle (13 months post infection), the milk of 22 lactating sheep was tested using the m-ELISA. Sera from the same sheep were also tested for Brucella antibodies using the Rose Bengal test (RBT) and the complement fixation test (CFT). The first serum sampling after parturition showed 100% sensitivity in both the RBT and the CFT (confidence interval [CI] 94-100%), but in subsequent samplings the sensitivity of the RBT decreased to 73% (CI 55-85%). Similarly, the sensitivity of the CFT decreased two months after the first sampling, when respective sensitivities of 95% (CI 81-98%) and 81% (CI 61-93%) were recorded for the final two samplings. The sensitivity of the m-ELISA decreased initially (68% on the third sampling, CI 50-81%), but then increased to 95% (CI 81-98%) for the final sampling. When disease prevalence in a flock is below 5%, the estimated probability of not detecting an infected flock through m-ELISA bulk milk testing is over 25%. Under field conditions in Italy (average sheep flock size of 70), the probability that the infection will not be detected is over 25% when four (or less) infected milking sheep are present in the flock. The results show that the m-ELISA is not a reliable screening test for bulk milk samples when the prevalence of brucellosis in a sheep flock is low.