The inference drawn from the ordinary multiple regression approach may be questionable when this method is used to analyze hydrologic data. A statistical model that avoids some of these uncertainties is developed. A Taylor series expansion is suggested to obtain exponential and interaction terms. Orthogonal transformations are used to extract some of the variables for use as predictors. A rule is exhibited for selecting the single most important (based on ability to explain variance in the dependent variable) independent variable, testing its significance, and removing its effects on all remaining variables. A second rule is exhibited for selecting, in turn, the succeeding most important independent variables, testing their significance, and removing their effects from other variables. Finally, a rule is exhibited for stopping the selection of independent variables when those remaining will not contribute significantly to the further reduction of unexplained variance in the dependent variable. The net result is the selection of a few from many independent variables to use in a ‘near best’ prediction equation. A method is presented for obtaining the multiple regression equation using only the selected variables. The application of the model is illustrated by its use in analyzing data from 763 storms on watershed 3H at the Central Great Plains Experimental Watershed, Hastings, Nebraska.