Water solubility is a critical property in risk assessments for chemicals, but measured values are often unavailable. The most practical approach to estimating solubility for diverse and complex structures involves regression against the logarithm of the octanol/water partition coefficient (log Kow). However, currently used equations were developed from relatively small training sets (200–300 compounds). Our objective was to upgrade log Kov-based methods by examining a much larger and diverse training set and, if possible, to improve estimation accuracy further by incorporating a series of simple correction factors. We also evaluated the contributions of melting temperature (Tm) and molecular weight (MW). Measured values of water solubility, log Kow and Tm (solids only) were available for all compounds in the training set (n = 1,450). The best equation for this dataset included log Kow, Tm, MW, and 12 correction factors as independent variables, with r2 = 0.970, SD = 0.409, ME = 0.313. Statistics for an independent validation set of 817 compounds were consistent with these results: r2 = 0.902, SD = 0.615, ME = 0.480. We found that Tm contributed significantly to estimation accuracy for solids; addition of MW also increased accuracy; measured values of log Kow and Tm produced more accurate solubility estimates than did estimated values; and the new method outperformed other widely used general-purpose equations based on log Kow.