Data from 42 lakes and reservoirs in Virginia, North Carolina, South Carolina, and Georgia were used to develop and evaluate cross-sectional regression models for phosphorus. For all models under consideration, the endogenous variable was lakewide average annual phosphorus concentration, and the exogenous variables were phosphorus input to the lake and various geomorphologic and hydrologic variables. A linear model was developed for logarithmically-transformed terms that resulted in a substantially lower prediction sum of squares than did other models under consideration. Among the other models examined were those specified and fitted to north temperate lakes by other investigators. Parameter characteristics and error terms for the proposed model were studied with particular attention directed to the a priori assumption of a strictly linear relation between input and concentration of phosphorus. The calculation of prediction uncertainty is illustrated in an application of the model using an empirical Bayes approach and a prior model based on linearity.