Currently we operate in a data rich environment that is under utilized for analysis. Better methods of analyzing this data are needed in order to extract useful information. This is especially true for predictive and diagnostic purposes. In many cases, the data can be made commensurate and thus can be simultaneously analyzed on all dimensions. The assumption is that like data points are closer together with respect to distance than unlike data points, one can use cluster analysis to find groups of similar points. These groups or clusters are used in the classification or sorting of unknown objects. Many existing methods of cluster analysis have limitations. They do not lend themselves to practical application without prior knowledge about the data in question. This includes the number of clusters, cluster shape, probability distribution functions, outlier points, etc. Some of these methods are unable to dynamically respond to newly collected data after initial clustering has taken place. Therefore they can’t adapt to new situations. A novel self-clustering algorithm that is not subject to the above limitations and that will find and define clusters in n-dimensional data space is described. One immediate application for this new self-clustering algorithm is to create statistical bin limits (SBL) at wafer sort. If a die fails at wafer sort, the fail is categorized into one of several bins. Using SBL, an upper limit is placed upon the bin. An SBL fail is triggered if the number of fails in any bin exceeds that bin’s SBL. A wafer with an SBL fail is considered to be suspect with respect to quality and reliability or may be symptomatic of a larger yield issue. This process can also be applied on a by lot basis.