At various steps of IC manufacturing, defects need to be inspected to ensure process integrity., In this article, we describe a new methodology for defect inspection: defect self-inspection, i.e., defect detection based on a captured optical image itself without comparing to a reference image, where the reference image is a database image or an optical image from another die. Through intuitive understanding of human intelligence in detecting defects, we propose a systematic methodology of self-inspection based on pattern code space and code vectors. We describe the general methodology and present its implementations for two common types of defects as specific examples: detection of edge intrusion and protrusion (also known as mouse-bite) and detection of particle or contamination (also known as pinhole/pin-dot). (C) 2004 American Vacuum Society.