In a typical standard costing or budgetary control system, a manager might receive a variance analysis reporting several hundred variances, of which many may have arisen due to random factors, or may be too insignificant to merit attention. The managers uses his knowledge and experience to identify the important variances which demand further investigation. Both management accounting theory and statistical decision theory can make significant contributions towards improving this decision, but both make extravagant demands on the manager in terms of the theoretical and factual knowledge expected of him. Much of this knowledge, even if available, is scattered throughout the organization with very little readily accessible to top management itself. This research takes the view that a knowledge. Base / Expert System approach can be useful in this context. An expert system was implemented in Turbo Prolog using fuzzy logic and MYCIN type certainty factors to handle uncertainty. Though traditional Prolog interpreters can be used directly to write Expert Systems without using an Expert System Shell, this is not the case with the Turbo Prolog compiler. It becomes necessary to write an interpreter/Expert System Shell in Turbo Prolog using some of the software tools (scanner and parser)available along with the Turbo Prolog compiler itself. The expert system was tried out on a case on cost variance investigation from a well known book on management control systems (Antony, Dearden and Bedford, 1984). The substantive performance of the system in this armchair case study was quite encouraging. In terms of speed and memory requirements, the system is close to the limits of what is possible in the PC environment with Turbo Prolog. It is likely that further work in this area will have to move out of the PCs to the workstations or to other more powerful computing platforms. The most important enhancement that is needed in the current system is a natural language interface; the current Prolog-like interface is acceptable only in classroom/research settings. The system has had considerable success in its principal research objectives. However, on the question of integrating statistical decision theory with fuzzy logic and certainty factors, the expert system methodology appears to be at a dead end; perhaps real progress in this area will come from purely statistical approaches to the problem.