Large multivariate data sets still challenge visualization techniques and their interactive capabilities. In this paper we present a new faster formulation of a high precision projection technique that allows much larger data sets to be handled and interacted with. By using this novel approach, named Piecewise Least Square Projection (P-LSP), we provide new and effective ways of exploring the data set through its feature space. The paper shows the use of P-LSP to help data selection by the user during exploration using coherence amongst data items. Examples are shown for images and for volumes as large as 1,000,000 voxels, with as many as 14 variables per voxel, although the technique can also be used for any data set for which a reasonable feature space can be determined.