A QSAR analysis was carried out on a dataset of 126 anthraquinone-based cytotoxic compounds. A PCA of the molecular descriptors was used to cluster the dataset into smaller subsets according to their structural features and QSAR models were derived for the selected sets. During the modeling, protonated states of molecules and nonlinear transformations of the descriptors were considered. The developed models have been interpreted in the context of cytotoxicity and validated with leave-one-out, and leavemany-out cross-validation. The descriptors in the resulting models describe the size and charge distribution of molecules although in different clusters their proportions vary.