Problem solving in the tax domain requires two kinds of knowledge: of the law itself and of how the law has been applied in the past. The need for the second factor arises as a result of the ambiguity of natural language. The problem solver requires information on how the courts have adjudicated specific cases in the past. This information would then provide the basis for reasoning about the current case. In this paper we address the issue of developing a system which will retrieve relevant historical cases. The cases are stored using a frame representation scheme and the users can retrieve cases by specifying either attributes alone or attributes and values associated with them. Currently the system has been implemented in Pascal on a Cray. The case base contains 250 cases relating to Section 183 of the tax code.
