Water quality models have been applied to the six University Lakes of Baton Rouge, La., to assist in the analysis of effects of restoration efforts. Phosphorus was identified as the nutrient limiting algal growth in these lakes. A relationship between total phosphorus and fishkills was identified. Initial applications of a modified Vollenweider model indicated that the model was capable of projecting long-term average lake conditions but that it was limited by its inability to represent short-term variations. These variations were crucial for the projection of fishkill episodes. Intensive surveys failed to identify practical modifications to the model within a deterministic format and a stochastic approach was undertaken. Uncertainties were partitioned into terms representing lumped modeling errors and sampling variability. This approach produced results that were suitable for interfacing with objective functions based upon total phosphorus levels. The need for more widespread application of stochastic techniques was evident from experiences associated with this project.