Empirical equations are frequently used to predict potential fish harvest from lakes or reservoirs. The prediction limits around these estimates have not been evaluated. We found that prediction limits from Ryder's morphoedaphic equation were inaccurate because of artificial correlation between dependent and independent variables used in regression. Furthermore there are inconsistencies between different versions of the model. Reanalysis of Ryder's data, gave upper 95% prediction limits that were 4.8–5.9 times the lower 95% prediction limit. Separating the component variables of the morphoedaphic index reduced prediction limits slightly. New upper limits were 4.5–5.6 times the lower 95% limits. Theoretical scaling of fish harvest on lake volume, consistent with the known tendency of small lakes to yield more fish per unit area, reduced the upper prediction limits to 2.8 times the lower limit. The model, which was allometric in form, correctly related the exponent relating harvest to lake volume in two independent data sets.