The first part of this paper demonstrates a simple graphical way to visualize estimated variances, in terms of a plot of the total initial variance ("SIGNAL") versus residual variance ("NOISE"), as a pragmatic alternative to tables of F-tests. The recently developed Procrustes rotation in the bilinear "jack-knifing" form is then presented as a method for simplifying the comparison of PLS Regression models from different data sets. These methods are applied to sensory data in order to study if naïve (untrained) sensory panelists can produce reliable descriptions of systematic differences between various test meals. The results confirm that three panels of 15 naïve assessors each could give repeatable intersubjective description of the most dominant sensory variation dimensions.