The paper discusses a strategy for polychotomous classification that involves estimating class probabilities for each pair of classes and then coupling these estimates together. The coupling model is similar to the Bradley-Terry method for paired comparisons. The authors study the nature of the class probability estimates and examine the performance of the procedure in real and simulated data sets. They consider various classifiers, including linear discriminants, nearest neighbors, adaptive nonlinear methods, and the support vector machine. The paper also explores the properties of the coupling solution and its relation to Friedman's "max-wins" rule, as well as the benefits of pairwise threshold optimization. The authors provide examples and simulations to demonstrate the effectiveness of the coupling procedure, showing that it often outperforms other methods in terms of classification accuracy.The paper discusses a strategy for polychotomous classification that involves estimating class probabilities for each pair of classes and then coupling these estimates together. The coupling model is similar to the Bradley-Terry method for paired comparisons. The authors study the nature of the class probability estimates and examine the performance of the procedure in real and simulated data sets. They consider various classifiers, including linear discriminants, nearest neighbors, adaptive nonlinear methods, and the support vector machine. The paper also explores the properties of the coupling solution and its relation to Friedman's "max-wins" rule, as well as the benefits of pairwise threshold optimization. The authors provide examples and simulations to demonstrate the effectiveness of the coupling procedure, showing that it often outperforms other methods in terms of classification accuracy.