This document describes a method called "k-means" for partitioning multivariate data into k groups based on minimizing within-group variance. It can be used for applications like classification, prediction, and testing independence among variables. The k-means process works by starting each group with a single data point, then iteratively assigning new points to the nearest group and updating group means. The document proves that the k-means values converge almost surely to a set of distinct points that minimize within-group variance. It also describes some applications and extensions of the k-means method.