The purpose of this study is to develop a new cross-peeling technique that can improve prediction accuracy and misclassification costs by using Data Envelopment Analysis (DEA) and Negative DEA together as methods for corporate bankruptcy prediction. To this end, this study applied four methods to bankruptcy prediction and compared their performance: three existing nonparametric approaches (the method using optimal cutoff points of DEA and Negative DEA, the method applying the peeling technique to Negative DEA, and Simak's peeling technique) and the cross-peeling technique developed in this study. The empirical analysis results showed that for the bankruptcy prediction method using optimal cutoff points of DEA and Negative DEA, the optimal cutoff points exhibited high variability depending on the time period of analysis, thus limiting the practical applicability of this technique. Furthermore, since the prediction accuracy of both the method applying the peeling technique to Negative DEA and Simak's peeling technique varies depending on the number of classification layers, it became apparent that finding the optimal number of classification layers to enhance prediction accuracy is a critical issue when using these two techniques for bankruptcy prediction. In contrast, the cross-peeling technique developed in this study was confirmed to yield improved results in terms of both prediction accuracy and misclassification costs compared to the two existing peeling techniques, without the need to find an appropriate number of classification layers.