If knowledge can be gained at the pre-processing stage, concerning the approximate underlying structure of large databases, it can be used to assist in performing various operations such as variable subset selection and model selection. In this paper we examine three methods, including two evolutionary methods for finding this approximate structure as quickly as possible. We describe two applications where the fast identification of correlation structure is essential and apply these three methods to the associated datasets. This automatic approach to the searching of approximate structure is useful in applications where domain specific knowledge is not readily available.