A total of 6289 drug-like anticancer molecules were collected from Binding database and were analyzed by using the classification techniques. The collected molecules were encoded to a diverse set of descriptors, spanning different physical and chemical properties of the molecules. A combination of genetic algorithms and counterpropagation artificial neural networks was used for navigating the generated drug-like chemical space and selecting the most relevant molecular descriptors. The proposed method was used for the classification of the molecules according to their therapeutic targets and activities. The selected molecular descriptors in this work define discrete areas in chemical space, which are mainly occupied by particular classes of anticancer molecules. The obtained structure-activity relationship (SAR) patterns and classification rules contain valuable information, which help to screen the large databases of compounds, more precisely. Such rules and patterns can be considered as virtual filters for mining the large databases of compounds and are useful in finding new anticancer candidates.