This paper proposes an algorithm for data mining named Pheromone-Miner (ant-colony-based data miner). The algorithm is inspired by both researches on the behavior of real ant colonies and data mining concepts as well as principles. The goal of Pheromone-Miner is to extract more exact knowledge from a database. Pheromone-based mining breaks through limitations of other mining approaches. We compare the performance of pheromone-miner with a general semantic miner. The accident causes discovered by ant-miner are considerably more accurate than those discovered by a general semantic miner. In a word, this evolutionary algorithm is suitable for improving the accuracy of data miners.