As part of our continuing effort to improve monitoring and control in the pulp and paper industry, this work is concerned with fault detection and diagnosis in the papermaking process. It analyzes on-line data gathered from the real industrial process, investigates the cause/effect relationships, identifies latent variables responsible for paper machine operation and paper quality, and constructs multivariate control charts for more effective supervision and control. Based on multivariate statistical analysis, a new diagnosis method using the directional property of the observation residuals is also developed.