Computer-aided design of new guanidinium salts was explored and experimentally tested, en route to the discovery of new ionic liquids. Quantitative structure-property relationships were established to predict the mp of guanidinium salts of four different anionic families (Cl-, BPh4 -, Br-, and I-). Models were built with a data set of 101 salts and counterpropagation neural networks. Predictions for an independent test set were obtained with R2=0.815, and a fivefold cross-validation procedure yielded R2=0.742. Assisted by the models, six new guanidinium salts were prepared, and the measured melting properties were reasonably in accordance with the predictions. One of the new chloride salts is liquid at room temperature, and three tetraphenylborate salts have mp values lower than those previously available in the data set for that anion.