Flood data were assembled for 168 Scottish basins containing 3071 station-years of record. Multiple regression techniques were used to produce equations for predicting mean annual flood from physiographic and climatological basin characteristics. Mean annual rainfall (SAAR) gave better results than measures of extreme rainfall (M52D and RSMD). Percentage area of lake storage (LOCH) was found to be a better predictor than fraction of the basin draining through a lake (LAKE). Apart from SAAR and LOCH, the recommended equation requires the basin area (AREA), stream frequency (STMFRQ) and an index of the soil type (SOIL). Measures of average basin slope (AVES) and main stream slope (S1085) were not statistically significant. The standard error of the estimate of the predicted mean annual flood is 0.147. The equation has a coefficient of determination, R2, of 0.914 and appears robust over a wide range of basin types without requiring regionally derived multipliers or coefficients. No significant improvement in fit was acheived by ridge regression, Stein estimation or use of the influence function.