Weed control efficiency and the environmental fate of herbicides depends on the interactions between the pesticide molecule and the soil. Knowledge of these factors is essential to increase weed control efficiency and reduce environmental contamination of these products. Two studies were carried out to evaluate the effects of soil attributes on sorption and desorption of hexazinone herbicide. In the first experiment the influence of pH by the addition of limestone on the sorption and desorption of hexazinone in 8 soles was analyzed. The second study evaluated the combination of multivariate techniques for the creation of multiple regression models to estimate the Kfs and Kfd values of soils for hexazinone, based on the chemical and physical attributes of the soil. The sorption evaluation was performed using the sorption isotherms with solutions of the herbicide (0.10, 0.22, 0.45, 1.00, 2.00, 3.50 and 7.00 mg L-1) prepared in 10 mmol L-1 CaCl2 and added with 8.0 mL to 4.00 g soil samples for stirring. The desorption was evaluated by the construction of the desorption isotherms from the collection of the supernatant from the tubes of the sorption tests, to which were added 8.0 mL of the 10 mmol L-1 CaCl2 solution free of herbicide. All the experiments were carried out in triplicates using a high performance liquid chromatography system. The limestone application did not alter the equilibrium sorption time of the herbicide for the same soil class. The values of Kfs and kfd of the hexazinone herbicide were higher in soils with higher content of organic matter and clay. In all soils the herbicide returned to the solution by desorption process, and that liming implied a reduction in Kfd, indicating a higher availability of hexazinone in the soil solution. Liming reduced sorption of the herbicide in all evaluated soils. The analysis of correlation and main components did not allow the generation of multiple linear regressions capable of estimating the coefficients Kfs and Kfd. The multiple linear regression technique generated models with greater predictability when associated with correlation analyzes and main components with cluster analysis. Regression models formed from groups of soils with greater similarity presented high adjustment and prediction power of Kfs and Kfd