<p>Ant colony optimization (ACO) is adopted extensively in feature selection on account of parallel computing and strong robustness. However, owing to slow rate of convergence and long search time, the current feature selection methods based on ACO cannot handle large-scale datasets well. For solving the problem, a feature selection algorithm based on ant colony optimization and mutual information (ACO-MI) is given. ACO-MI first adopts mutual information for measuring the relevance and removes irrelevant features. It uses the method of equal interval division and ranking to handle mutual information and constructs similarity matrix. Then, it utilizes the number of features and detection result of classifier to construct the memory based on fitness. It updates the pheromone density and combines similarity matrix to determine the optimal features’ interval and the optimal features. To verify the effect of ACO-MI, several typical feature selection algorithms are compared. Experimental results show that ACO-MI obtains better feature selection performance.</p>