An equation modeling on Sembulan river, Sabah, Malaysia, has been undertaken using a backward stepwise multiple linear regression. A good performance has been obtained using a log transformation on water quality data designated as predictors and dependent variable. The regression model is in accordance with the ANOVA result. The temperature, biochemical oxygen demand (BOD), Echerichia Coli, Pb and nitrate were described as continuous predictors, while the river location (downstream, municipal and upstream) was designated as independent string grouping variable, and the chemical oxygen demand (COD) was set up as the dependent variable. The string grouping variable was converted to its dummy variable, which in turn led to design of a three-equation model with respect to river location. The results show that BOD has a strong effect on COD, while Pb and nitrate show less effect on COD. The temperature gives little negative effect on COD, while other variables such as pH, salinity and Cd are excluded from the river modeling since they induce insignificant effects based on backward criterion probability of F-value ≥ 0.100. Using the general linear model with LSD mode, it is revealed that predictor(s) show a remarkable discriminant effect between upstream and municipal/downstream on the 0.05 level. The most effect came from salinity indicated by the canonical discriminant function based on Wilks’ lambda.