This article reports an empirical investigation of the accuracy of rules that classify examples on the basis of a single attribute. On most datasets studied, the best of these very simple rules is as accurate as the rules induced by the majority of machine learning systems. The article explores the implications of this finding for machine learning research and applications. The study compares the performance of 1R, a program that learns 1-rules from examples, with C4, a state-of-the-art learning algorithm, on 16 commonly used datasets. The results show that 1R's accuracy is only slightly lower than C4's on most datasets. On average, 1R's accuracy is 5.7 percentage points lower than C4's, but on 12 of the 16 datasets, the difference is less than the average. On 14 of the 16 datasets, 1R's accuracy is an average of 3.1 percentage points lower than C4's. On half the datasets, 1R's accuracy is within 2.6 percentage points of C4's. The study also examines an upper bound on the accuracy that could be achieved by improving 1R's selection criterion. This upper bound, called 1R*, is very similar to the accuracy of C4's decision trees. The results suggest that 1R could be competitive with C4 if its selection criterion is improved, but it would never significantly outperform C4. The study also explores the use of 1-rules to predict the accuracy of complex rules. The results show that 1Rw, the accuracy of 1-rules when the whole dataset is used for both training and testing, is a good predictor of C4's accuracy on most datasets. 1Rw is highly correlated with the median accuracy of the datasets, and it can be used to predict the accuracy of other machine learning systems. The practical significance of these results is discussed, with the conclusion that very simple rules often perform well on most commonly used datasets. This suggests that simple-rule learning systems are often a viable alternative to systems that learn more complex rules. The study also highlights the importance of considering the practical significance of datasets in machine learning research.This article reports an empirical investigation of the accuracy of rules that classify examples on the basis of a single attribute. On most datasets studied, the best of these very simple rules is as accurate as the rules induced by the majority of machine learning systems. The article explores the implications of this finding for machine learning research and applications. The study compares the performance of 1R, a program that learns 1-rules from examples, with C4, a state-of-the-art learning algorithm, on 16 commonly used datasets. The results show that 1R's accuracy is only slightly lower than C4's on most datasets. On average, 1R's accuracy is 5.7 percentage points lower than C4's, but on 12 of the 16 datasets, the difference is less than the average. On 14 of the 16 datasets, 1R's accuracy is an average of 3.1 percentage points lower than C4's. On half the datasets, 1R's accuracy is within 2.6 percentage points of C4's. The study also examines an upper bound on the accuracy that could be achieved by improving 1R's selection criterion. This upper bound, called 1R*, is very similar to the accuracy of C4's decision trees. The results suggest that 1R could be competitive with C4 if its selection criterion is improved, but it would never significantly outperform C4. The study also explores the use of 1-rules to predict the accuracy of complex rules. The results show that 1Rw, the accuracy of 1-rules when the whole dataset is used for both training and testing, is a good predictor of C4's accuracy on most datasets. 1Rw is highly correlated with the median accuracy of the datasets, and it can be used to predict the accuracy of other machine learning systems. The practical significance of these results is discussed, with the conclusion that very simple rules often perform well on most commonly used datasets. This suggests that simple-rule learning systems are often a viable alternative to systems that learn more complex rules. The study also highlights the importance of considering the practical significance of datasets in machine learning research.