Abstract: Unknown empirical parameters in water retention models may reduce the reliability of prediction. The cube-shaped assemblages of natural soil and idealized spherical particles are developed according to the fractions from particle-size distribution. On the basis of geometry and physical properties, a relationship of proportionality between them is then proposed to provide a physical approach for computing pore radii and water retention curves of soils from particle-size fractions, bulk density and particle density without incorporating unknown empirical parameters. Finally, the physical approach is validated against the test data of water retention for a total of forty soil samples from the hydraulic property database UNSODA. The results of RMSR between the predicted and the measured values of matric suction show that the distribution of RMSR values ranges from 0.179 to 0.833. This physical approach is also superior to the traditional Arya model which requires two unknown empirical parameters for describing the water retention curves for thirty seven of the forty soil samples.