Paratuberculosis (Johne's disease) in ruminants is caused by Mycobacterium avium subsp. paratuberculosis (MAP). Owing to the lack of accurate laboratory tests, diagnosis is challenging in subclinically infected cattle. To evaluate the long-term performance of serum ELISAs for the detection of paratuberculosis in a dairy herd with low MAP-prevalence, three investigations of all the cows and the consecutive testing of 33 cows suspected to be infected with MAP and 30 cows classified as MAP free were performed over a period of 22 months. Blood samples were tested by three commercial serum ELISAs, MAP shedding was detected by bacteriological culture and polymerase chain reaction (PCR). The ELISA results varied in a wide range in the herd investigations with 1.2% to 18.8% positive samples, the faecal samples were positive for MAP between 1.8% and 4.9% in the three herd investigations. Over the study period, ELISA-positive serum samples varied between 0.0% and 69.7% in MAP-suspicious and 0.0% and 17.6% in MAP-unsuspicious cows with a poor correlation between ELISAs and faecal shedding. The correlation coefficient of the optical density values of the three ELISAs varied between 0.348 and 0.61. Evidence of cow specific variations of residuals was found in all linear models. The linear mixed models showed relevant contribution of cow specific variation in explanation of the residual variances. They also showed significant effects of the explanatory ELISA, the group (MAP-suspicious or MAP-unsuspicious) and the time of sampling. It can be concluded that the choice of the laboratory test significantly influences the outcome of the testing for MAP and that none of the three ELISAs can be thoroughly recommended as single test for the early diagnosis of paratuberculosis in cattle. Test results should always be interpreted with caution to avoid erroneous decisions and the disappointment of those engaged in the abatement of paratuberculosis.