An ELISA containing lipoarabinomannan (LAM) antigen was used to detect antibodies in milk and serum for diagnosis of Mycobacterium paratuberculosis infection in dairy cattle. In experiment 1, milk and serum samples were obtained from 25 cows, and subjected to LAM ELISA testing immediately, and after 1 year of storage at -70 C. Milk samples, with and without a commonly used chemical preservative, were tested. There was no significant difference in LAM ELISA results between fresh and frozen samples or between preserved and unpreserved milk samples. In experiment 2, milk samples were collected daily from 30 cows over a 14-day period. The day-to-day coefficient of variation was 0.19 for milk LAM ELISA and was 0.15 for serum LAM ELISA, with no statistically significant time effect detected. In experiment 3, single milk, serum, and fecal samples were obtained from 764 cows. The fecal samples were cultured for M paratuberculosis to identify infected cows, and the serum and milk samples were subjected to LAM ELISA testing. Results were compared, using the area under the receiver operating characteristic curves. The milk LAM ELISA had specificity (+/- 95% confidence limits) of 87 +/- 8.1% when the cutoff was set at 50% sensitivity, and specificity of 83 +/- 9.1% when sensitivity was set at 60%. The area under the receiver operating characteristic curve was 0.85 +/- 0.03 for the milk ELISA and 0.75 +/- 0.02 for the serum ELISA. In this population of cattle, the milk LAM ELISA had comparable accuracy to serum LAM ELISA, although the milk LAM ELISA was slightly less reproducible (higher coefficient of variation).