Antioxidative activity expressed as protection of thymidine has been investigated for a set of 30 pulvinic acid derivatives. A combination of in vitro testing and in silico modeling was used for synthesis of new potential antioxidants. Experimental data obtained from a primary screening test based on oxidation under Fenton conditions and by an UV exposure followed by back-titration of the amount of thymidine remaining intact have been used to develop a computer model for prediction of antioxidant activity. Structural descriptors of 30 compounds tested for their thymidine protection activity were calculated in order to define the structure-property relationship and to construct predictive models. Due to the potential nonlinearity, the counter-propagation artificial neural networks were assessed for modeling of the antioxidant activity of these compounds. The optimized model was challenged with 80 new molecules not present in the initial training set. The compounds with the highest predicted antioxidant activity were considered for synthesis. Among the predicted structures, some coumarine derivatives appeared to be especially interesting. One of them was synthesized and tested on in vitro assays and showed some antioxidant and radioprotective activities, which turned out as a promising lead toward more potent antioxidants.