This paper introduces nonparametric Bayesian models, in particular, Dirichlet process mixture (DPM) models and the infinite relational model (IRM) as an extension of DPM for multi-way data clustering. The nonparametric Bayesian modelling is a more flexible approach than a standard parametric Bayesian modelling in that a nonparametric prior distribution over model parameters is incorporated into the data gereration process. More specifically, DPM enables us to define distributions over the countably infinite sets by exploring their clustering structures. In this paper, we explain the basic idea of DPM modelling and its learning algorithms. We also illustrate practical usefulness of DPM modelling though experimental results using IRM.