For effective irrigation management we need to know the water storage capacity of the soil reservoir. Though plant extractable water is best measured in the field, sometimes it is useful to estimate it. Laboratory-derived retention curves do not necessarily reflect field conditions. Statistical models to estimate plant extractable water from other soil properties are restricted by assumptions that are difficult to check, and they can look very complicated. We propose to test a physical-based model that exploits the similarity between the particle size distribution curve and the soil water retention curve. A large data set of soil properties from the USA was used. Detailed particle size fraction data enabled the construction of simulated soil water retention curves for 388 samples. The physical-based model was compared against a statistical model that was derived from a subset of the data base. The statistical model fit the data better than the physical-based model. On the other hand, the statistical model overpredicted the soil water limits of those soils that were not used in the derivation of the statistical model. The strength of the physical model is that it represents a cause and effect relationship between particle size distribution and soil water retention. Also, it is conceptually simple and requires few inputs. The physical model may be improved by considering soil structure and type of clay.