Can. J. Fish. Aquat. Sci. M(Suppl. 2 ) : 471485.Methods for estimating fish production in aquatic ecosystems range from simple empirically derived estimators, such as morphoedaphic indices, to complex ecosystem simulation models. As first-order estimators, the former are attractive to managers because they are simple and relatively inexpensive to apply and interpret. Application of the latter group has been limited because many of the data inputs are difficuit and expensive to obtain. Between these extremes are several models, such as the biomasssize spectrum model, that provide useful information for moderate expenditures of time and effort. Existing and new methods are reviewed in the light of production theory and several are applied to Great Lakes and Lake Winnipeg data. Eight empirical models derived from limnological variables were selected from the literature and used to estimate potential fish yield for the Great Lakes and Lake Winnipeg. The models predicted a fairly narrow range of potential yields, but when compared with historic yields, none was consistent for all lakes. The best overall empirically derived estimator of potential yield in the Great Lakes was the morphoedaphic index. Potential fish production estimated from invertebrate production with Borgmann's biomasssize spectrum mode! was considerably greater than historic yields or the yield theorie de la productivite et on applique piusieurs de ces m6thodes B des donn6es recueillies dans \es Grands Lacs et E e lac Winnipeg. On a choisi dans les ouvrages publies huit modeles empiriques tires de variables limnologiques; ils sewent a determiner le rendement potentiel en poissons des Grands Lacs et du lac Winnipeg. Les modeles predisent une gamme assez petite de rendements potentiels; toutefois, une '~oncribution No. 86-03 sf the Ontario Ministry of Natural Resources, Research Section, Fisheries Branch. Box 50, Maple, Ont. LOJ fE0. Can. J. Fish. Aquat. Sci. Downloaded from www.nrcresearchpress.com by Duke University on 10/14/12 For personal use only.Since Ryder's (1965) landmark paper there has k e n extensive Base of ME1 in addition to other physical, chemical, and biological indices. In general, such indices are designed for simplicity of application. They treat yield of fish as a "shadow" statistic representing fish production. This embodies the implicit assumption that the yield for a given lake represents the same fraction of total production as the yields for the other lakes with which it is to be compared. Where reliable and comparable