A random sample of continuous measurements can be partitioned into g groups or clusters by minimizing the within group dispersion as measured by the 1-norm. The central limit theory associated with such partitions which are universally optimal or locally optimal is derived. A procedure is...