The traditional approach to statistical process control, as developed by Shewhart in the 1920's, focuses on process monitoring and change-point detection. There has been little emphasis on the development of statistical methods for failure diagnostics and process improvement. There is a critical need for such methods in advanced manufacturing industries where a considerable amount of in-process and product quality data are now being collected routinely. These data contain useful information that can be exploited for improving the processes. This talk will outline a statistical framework for intelligent manufacturing and use some real applications from the automotive and semiconductor industries to describe the research issues. In particular, methods for identifying spatial patterns of defects will be discussed.