Accuracy is a primary concern in all applications. The more accurate classification trees we generate, the more useful they would be. This paper concerns about the improvement in terms of classification accuracy. The author has proposed Class Proportion Technique (CPT) to split the original data set into two disjoint sets for training and testing purpose. The results of the study indicate that CPT is better than other techniques. Accuracies have been compared with the existing well known researcher’s results.