Background Small molecules that bind reversibly to DNA are among the antitumor drugs currently used in chemotherapy. In the pursuit of a more rational approach to cancer chemotherapy based upon these molecules, it is necessary to exploit the interdependency between DNA-binding affinity, sequence selectivity and cytotoxicity. For drugs binding noncovalently to DNA, it is worth exploring whether molecular descriptors, such as their molecular weight or the number of potential hydrogen acceptors/donors, can account for their DNA-binding affinity and cytotoxicity. Results Fifteen antitumor agents, which are in clinical use or being evaluated as part of the National Cancer Institute's drug screening effort, were analyzed in silico to assess the contribution of various molecular descriptors to their DNA-binding affinity, and the capacity of the descriptors and DNA-binding constants for predicting cell cytotoxicity. Equations to predict drug-DNA binding constants and growth-inhibitory concentrations were obtained by multiple regression following rigorous statistical procedures. Conclusion For drugs binding reversibly to DNA, both their strength of binding and their cytoxicity are fairly predicted from molecular descriptors by using multiple regression methods. The equations derived may be useful for rational drug design. The results obtained agree with that compounds more active across the National Cancer Institute's 60-cell line data set tend to have common structural features.