High throughput techniques have found widespread application in pharmaceutical research and development. In the design of crystallization in particular, measuring solubility in a few dozen solvents is a commonly performed screening procedure to identify optimal solvent systems. However, due to the enormous potential combination of binary and ternary systems that may be employed, a purely empirical approach to optimize the crystallization solvent composition is intractable. Solubility modeling techniques are typically used to aid crystallization design by estimating the solubility of interest in untested systems. NRTL-SAC has proved a powerful tool to estimate solubility by modeling the activity coefficient of the solute with four adjustable parameters which are fitted from experimental data. Typically five to ten experimental values are sufficient to obtain the universal solute descriptors of the NRTL-SAC model. We present in this paper a systematic approach to optimize the data set used to fit and validate the NRTL-SAC parameters from experimental solubility obtained with high throughput technology.