In this paper we develop statistical models for the prediction of bankruptcy in Italian firms in the limited liability sector using annual balance-sheet information. Several issues are investigated, such as the structure of the data, the sampling procedure and the selection of financial predictors. Particular attention is given to the variable selection problem by comparing modern techniques, based on shrinkage, with traditional stepwise methods. The proposed models have been validated by means of accuracy measures over different time horizons. The results obtained give evidence in favor of the effectiveness of the proposed approaches at improving the models' predictive performance.