The most widely-used constitutive relations for unsaturated soils use matric soil suction as a state variable. The soil-water characteristic curve (SWCC), the relationship between soil suction and some measure of the water content, can be measured or predicted based on soil index properties such as the grain-size distribution (GSD) function. Estimation based on index properties is highly desirable due to its simplicity and low cost and would be the path of choice to the SWCC, provided the accuracy of the estimate were adequate. Whether measured or estimated, there is variability and uncertainty associated with the SWCC that in turn will directly impact any model for the unsaturated soil behavior that makes use of this relationship. The variability in the SWCC associated with direct suction measurements as well as the variability associated with the prediction of the SWCC based on GSD was investigated. Three different soils that cover a typical range of soils encountered in practice were used in this study. The investigated sources of variability related to direct suction measurements included the equations used to fit the suction data, the different methods available to measure suction, the operator, the number of data points used to define the SWCC, and the range of suction covered in the measurements. The variability observed in the SWCC when different predictive algorithms are used is also reported. Finally, a new model for predicting the SWCC based on soil index properties is presented, using a database of 190 soils. Results showed that the operator as well as the range of suction covered in the measurements produced significant variability in the measured SWCC. Surprisingly, the variability in the SWCC as predicted by soil index properties and/or GSD-based algorithms was found to be as small or smaller than that associated with the operator (person(s) measuring the SWCC).