Data mining offers an effective solution to product and process diagnosis in that it requires no additional equipment or capital investment and it does not cause interruption to production and operations. The most important role of data mining is the ability to separate data into different classes so that an accurate model can be built for fault detection, recognition and prediction. The proposed data mining methodology consists of data separability, hidden projection, back mapping, feature selection and reduction, and model building. The solution can be quantitative or qualitative depending on the pattern of the original data set. When combined with other soft computing techniques, data mining method can provide Application examples are briefly described to demonstrate the efficacy of the new method in diagnosis.