A **mixture model** is essentially an extension of the idea behind fuzzy K-means; however, it makes an assumption that there is an underlying probability distribution that generates the data. For example, we might assume that the data points are drawn from a set of K-independent Gaussian (normal) probability distributions. The cluster assignments are also soft, so each point is represented by *K* membership weights in each of the *K* underlying probability distributions.

See http://en.wikipedia.org/wiki/Mixture_model for further details and for a mathematical treatment of mixture models.