Multidimensional scaling is the general term for the set of approaches which share the aim of inserting odor stimuli, as points, into a geometrical space of low dimensionality (2-3 dimensions).
The properties of this space and of the points (which are the odors) are as follows:
(a) Stimuli which are close together, in terms of distance from each other in the space, are those which are perceptually quite similar to each other. Conversely, stimuli which are far away from each other are those which are perceptually quite dissimilar.
(b) In order to arrive at the proper positioning of points in this space, empirical estimates of overall qualitative dissimilarity have to be made for points of odors, The panelist (novice or expert) sniffs pairs of odorants and assigns a number to reflect how different they smell. These numbers then reflect subjective estimates of dissimilarity. The matrix of data that emerges fmm the study is triangular. The columns represent the odors, the rows represent the odors, and the entries represent the judgment of how different in odor quality each pair of odorants seems to be.