Significant progress has been made recently in the design of optimization models to mimic the decision making process of reservoir operators. Several of these models for real-time, short term reservoir system operation are described. The models include: an optimization model constructed specifically for a multi-reservoir, multipurpose system in Kentucky; a goal programming model that requires less information than most other models yet may yield equivalent or better results; a chance constrained model that explicitly considers errors in streamflow forecasts in the operating environment; and a balancing model that uses cumulative distribution functions to maintain a balance between storage, release, and any other characteristic of the operation. Each model is briefly described and comparisons between models are made. These models represent the initial development of an expert system for real-time reservoir system operation.