This research explores the theoretical and computational properties of a formal system that subsumes many of the inference mechanisms currently of interest in artificial intelligence research and applications. The theoretical work proposed involves developing a uniform framework of which various formal extensions to first order logic are special cases. The extensions we will consider include default reasoning, probabilistic inference, and reason maintenance systems. The research will enable us to recognize computationally effective (or ineffective) inference schemes in advance. We also intend to describe in a precise way the calculational compromises made by production systems such as MYCIN in implementing such extensions. Finally, we propose to implement a general-purpose inference engine incorporating the theory developed and capable of being conveniently specialized to perform inference suitable for any of a variety of domains. The significance of this project is that it will provide a bridge for artificial intelligence research between theoretical work on the nature of reasoning and practical work in developing programs such as expert systems. The benefits will include not only new theoretical insights but also new algorithms and tools which will allow more flexible development of expert systems, and improve the performance of such systems when they have been completed.