Multidimensional scaling is the term used to describe any procedure which starts with the 'distances' between a set of points, or information about these 'distances', and finds a configuration of the points, preferably in a small number of dimensions, usually 2 or 3. This type of approach is very useful because many sets of data arise, directly or indirectly, not as a (n × p) data matrix, but as an (n × n) matrix whose elements compare all pairs of individuals either by measuring 'similarity' or 'dissimilarity'. The two main types of scaling procedure are called classical scaling and ordinal scaling. This chapter discusses the many different measures which are available for assessing similarity and dissimilarity. Classical scaling is an algebraic reconstruction method for finding a configuration of points from the dissimilarities between the points, which is particularly appropriate when the dissimilarities are, exactly or approximately, Euclidean distances.