Abstract In many real-world classification problems, while there is a large amount of unlabeled data, labeled data is usually hard to acquire. One way to solve these problems is active learning. It aims to select the most valuable instances for labeling and construct a superior classifier. Most existing active learning algorithms are designed for binary classification problems, only a few algorithms can deal with multi-class cases. Moreover, as most multi-class active learning methods are directly extended from binary active learning methods, it is difficult for them to fuse the output results of binary cases. In this paper, we propose a novel multi-class active learning algorithm to tackle the above problems and select the most informative instances, called active learning with error-correcting output codes (ECOCAL). We create a codeword for each class and then obtain a test code for each unlabeled instance by error-correcting output codes (ECOC) framework, which is a powerful tool to combine multiple binary classifiers to address multi-class classification problems. By calculating the variance of the distance between a test code and all codewords, the proposed algorithm is able to measure the uncertainty across multiple classes. Extensive experimental results show that the proposed method outperforms several state-of-the-art active learning methods on both binary and multi-class datasets.