This paper presents a system that reasons from first principles, using knowledge of structure and behavior to diagnose digital electronic circuits. The system has been implemented and tested on several examples, demonstrating several advantages, including device independence, the ability to constrain initial hypotheses, and the capability to handle novel situations. The authors review their approach to describing structure and behavior, highlighting the use of "paths of causal interaction" and the concept of "adjacency" in understanding fault diagnosis. They introduce a technique called "constraint suspension," which helps determine which components can cause observed symptoms. The paper also discusses the importance of explicitly making assumptions underlying reasoning and provides a method for systematically enumerating these assumptions. The overall strategy for troubleshooting is based on the progressive relaxation of underlying assumptions, allowing the system to focus initially while expanding its scope to include a broad range of faults. The concept of adjacency is found to be useful in understanding why some faults are particularly difficult and why multiple representations can be beneficial. The paper concludes with a discussion of the limitations and future work, comparing their approach to previous methods in hardware diagnosis and fault models.This paper presents a system that reasons from first principles, using knowledge of structure and behavior to diagnose digital electronic circuits. The system has been implemented and tested on several examples, demonstrating several advantages, including device independence, the ability to constrain initial hypotheses, and the capability to handle novel situations. The authors review their approach to describing structure and behavior, highlighting the use of "paths of causal interaction" and the concept of "adjacency" in understanding fault diagnosis. They introduce a technique called "constraint suspension," which helps determine which components can cause observed symptoms. The paper also discusses the importance of explicitly making assumptions underlying reasoning and provides a method for systematically enumerating these assumptions. The overall strategy for troubleshooting is based on the progressive relaxation of underlying assumptions, allowing the system to focus initially while expanding its scope to include a broad range of faults. The concept of adjacency is found to be useful in understanding why some faults are particularly difficult and why multiple representations can be beneficial. The paper concludes with a discussion of the limitations and future work, comparing their approach to previous methods in hardware diagnosis and fault models.