Sun presents analysis, theoretical foundations, implementation architecture, and evaluation for a framework for reasoning with rules and similarities. The framework allows its users to deal with reasoning situations involving partial information, uncertain or fuzzy information, similarity matching, combinatorial rule interactions, and top-down and bottom-up inheritance. Analysis of the problem is based on a number of cases in human plausible reasoning. The theory of robust reasoning gives a unified approach to reasoning with rules and similarities. A two-level connectionist architecture is proposed as a computational mechanism for carrying out the theory. Evaluation involves analysis of the examples, evaluation of a number of reasoning patterns, and verification of the properties of the reasoning process. The framework may present a step forward in solving the brittleness problem of current rule-based systems. (The system is not able to deal with situations outside of its precise area of expertise.)