The aqueous solubility of poorly water-soluble drugs is an important property of many factors affecting their bioavailability such as the solubility and rate of dissolution in water. The quantitative structure-property relationship approach using genetic algorithm was applied to make models for predicting some poorly water-soluble drugs such as ursodeoxycholic acid, diphenyl hydrantoin and biphenyl dimethyl dicarboxylate. The experimental solubility data of 3518 chemical structures were collected from the web and used to build a model. Three data sets of 50 compounds were extracted according to their structural similarity with each drug. A fast and predictive similarity based approach was developed and validated with conventional method. This can be used to predict the aqueous solubility for drugs by using a small set of compounds, especially for poorly water-soluble compounds. Moreover, the estimation values of various sets were further compared with fine grinding experiment data.