One goal for eutrophication research is to predict lake trophic state without extensive field programs. However, existing comparisons of predicted:observed TP (total phosphorus) concentration calculate the former from measured watershed characteristics that require field sampling. Predictions of lake TP based on estimated rather than measured components are probably much less powerful, but this possibility should be quantitatively assessed. To assess the predictive power of current mass-balance models of eutrophication, measured estimates of volumetric water discharge (Qobs: m3·yr−1, n=110) and phosphorus load (Lobs: mg·m−2·yr−1, n=96) from lakes in the OECD (Organization for Economic and Cooperative Development) eutrophication data set were compared to calculated values of Q (Qcal) and 16 different variants of calculated phosphorus loading (Lcal, respectively. For the entire data set, Qcalwas unbiased and strongly correlated with Qobs. In regional comparisons, some Lcalvariants correlated well with Lobs, but others did not, and no single variant was significantly better than all of the others. In these comparisons, up to 95% of the variance in Lobswas explained by a single regional Lcalvariant, but the best regression of a single variant against Lobsfor the entire data set explained only 58% of its variance. Stepwise multiple regression to improve prediction of Lobsproduced the following model; Log Lobs= 1.23 + 0.81 (log Lbest) + 0.19(G) − 0.42(%F) − 0.38(%U)− 0.11 (log Ad); R2= 0.75,Sxy= 0.37, P < 0.001