Cancer is the leading cause of death among men and women under age 85. Every year, millions of individuals are diagnosed with cancer. But finding new drugs is a complex, expensive, and very time-consuming task. Over the past decade, the cancer research community has begun to address the in silico modeling approaches, such as Quantitative Structure-Activity Relationships (QSAR), as an important alternative tool for targeting potential anticancer drugs. With the compilation of a large dataset of nucleosides synthesized in our laboratories, or elsewhere, and tested in a single cytotoxic assay under the same experimental conditions, we recognized a unique opportunity to attempt to build predictive QSAR models. Early efforts with 2D classification models built from part of this dataset were very encouraging. Here we report a further detailed evaluation of classification models to flag potential anticancer activities derived from a variety of 3D molecular representations. A quantitative 3D-model model that discriminates anticancer compounds from the inactive ones was attained, which allowed the correct classification of 82 % of compounds in such a large and diverse dataset, with only 5 % of false inactives and 11 % of false actives. The model developed here was then used to select and design a new series of nucleosides, by classifying beforehand them as active/inactive anticancer compounds. From the compounds so designed, 22 were synthesized and evaluated for their inhibitory effects on the proliferation of murine leukemia cells (L1210/0), of which 86 % were well-classified as active or inactive, and only two were false actives, corroborating the good predictive ability of the present discriminant model. The results of this study thus provide a valuable tool for the design of novel potent anticancer nucleoside analogues.