The objectives of this research were to test the potential of unsupervised (USNN) and supervised neural network (SNN) models for detecting major and minor mastitis pathogens based on changes in milk parameters. A data set of 4,852 quarter milk samples with records for milk parameters and bacteriological status was used to train and validate the models by classifying milk samples into 3 different bacteriological states: not infected, intramammary infection (IMI) by minor pathogens, and IMI by major pathogens. Sensitivity of the USNN model was 97% for detecting noninfected quarters, 89% for minor pathogen IMI, and 80% for major pathogen IMI. Specificities of USNN models were close to 99% for all bacteriological states. The sensitivity of SNN models was affected by the ratio of infected to noninfected cases in the data set. As the ratio of infected to healthy cases increased from 1:1 to 1:10, detection accuracy for noninfected quarters increased from 82 to 98% but that for minor pathogen IMI decreased from 86 to 44%. The sensitivity for major pathogen IMI was 20% when the ratio was 1:1, but ranged from 20 to 40% when different ratios were tested. The SNN models indicated that somatic cell score and electrical resistance index had the most discriminating power. It was concluded that both USNN and SNN models were able to effectively differentiate between noninfected quarters and those infected by minor mastitis pathogens, and that the USNN model had a better agreement with results obtained from conventional microbiological methods. These types of models can be used in in-line milking systems to detect the infection status of a quarter and provide the farmer with diagnosing options for managing mastitis.