Density estimation is a central problem in data mining and knowledge discovery, with applications from data visualization and exploratory data analysis to supervised and unsupervised concept learning. This paper presents a simple nonparametric method for univariate density estimation that uses Bayesian inference and the minimum-message length principle to induce appropriate mixture models. Its advantage over existing methods is that it does not depend on ad hoc smoothing parameters or roughness penalties. The algorithm is efficient and robust.