Supervised learning consists of a large variety of methods that explore data relationships. The techniques described in this paper cover those methods that are robust and relevant to semiconductor data, sufficiently simple for use by non-statisticians, and proven effective in yield modeling. We first apply the classification and regression tree (CART) technique to detect the source of yield variations from electrical parameters and process equipment. Yield prediction models, including multinomial logistic regression (MNL) and the random forest (RF) method, will also be discussed. Case studies demonstrate the strength of combining traditional regression with machine learning techniques.