The sensitivity of current diagnostics for Johne's disease, a slow, progressing enteritis in ruminants caused by Mycobacterium avium subsp. paratuberculosis (MAP), is too low to reliably detect all infected animals in the subclinical stage. The objective was to identify individual metabolites or metabolite profiles that could be used as biomarkers of early MAP infection in ruminants. In a monthly follow-up for 17 months, calves infected at 2 weeks of age were compared with aged-matched controls. Sera from all animals were analyzed by .sup.1 H nuclear magnetic resonance spectrometry. Spectra were acquired, processed, and quantified for analysis. The concentration of many metabolites changed over time in all calves, but some metabolites only changed over time in either infected or non-infected groups and the change in others was impacted by the infection. Hierarchical multivariate statistical analysis achieved best separation between groups between 300 and 400 days after infection. Therefore, a cross-sectional comparison between 1-year-old calves experimentally infected at various ages with either a high- or a low-dose and age-matched non-infected controls was performed. Orthogonal Projection to Latent Structures Discriminant Analysis (OPLS DA) yielded distinct separation of non-infected from infected cattle, regardless of dose and time (3, 6, 9 or 12 months) after infection. Receiver Operating Curves demonstrated that constructed models were high quality. Increased isobutyrate in the infected cattle was the most important agreement between the longitudinal and cross-sectional analysis. In general, high- and low-dose cattle responded similarly to infection. Differences in acetone, citrate, glycerol and iso-butyrate concentrations indicated energy shortages and increased fat metabolism in infected cattle, whereas changes in urea and several amino acids (AA), including the branched chain AA, indicated increased protein turnover. In conclusion, metabolomics was a sensitive method for detecting MAP infection much sooner than with current diagnostic methods, with individual metabolites significantly distinguishing infected from non-infected cattle.