Use Normal Mixtures for clustering when your data come from overlapping normal distributions. If you have multivariate outliers, use the Robust Normal Mixtures method. You need to specify the number of clusters in advance.

Normal mixtures is an iterative technique based on the assumption that the joint probability distribution of the observations is approximated using a mixture of multivariate normal distributions. These mixtures represent different clusters. The individual clusters have multivariate normal distributions.

When clusters are well separated, hierarchical and k-means clustering work well. But when clusters overlap, normal mixtures provides a better alternative, because it is based on cluster membership probabilities, rather than arbitrary cluster assignments based on borders.