In this work, two reliable aqueous solubility models, ASMS (aqueous solubility based on molecular surface) and ASMS-LOGP (aqueous solubility based on molecular surface using ClogP as a descriptor), were constructed by using atom type classified solvent accessible surface areas and several molecular descriptors for a diverse data set of 1708 molecules. For ASMS (without using ClogP as a descriptor), the leave-one-out q(2) and root-mean-square error (RMSE) were 0.872 and 0.748 log unit, respectively. ASMS-LOGP was slightly better than ASMS (q(2) = 0.886, RMSE = 0.705). Both models were extensively validated by three cross-validation tests and encouraging predictability was achieved. High throughput aqueous solubility prediction was conducted for a number of data sets extracted from several widely used databases. We found that real drugs are about 20-fold more soluble than the so-called druglike molecules in the ZINC database, which have no violation of Lipinski's "Rule of 5" at all. Specifically, oral drugs are about 16-fold more soluble, while injection drugs are 50-60-fold more soluble. If the criterion of a molecule to be soluble is set to -5 log unit, about 85% of real drugs are predicted as soluble; in contrast only 50% of druglike molecules in ZINC are soluble. We concluded that the two models could be served as a rule in druglike analysis and an efficient filter in prioritizing compound libraries prior to high throughput screenings (HTS).