This Paper examines the file-merging technique of statistical matching. Details are given of both its practical application and its numerical accuracy. In particular, it is shown that the ability of statistical matching to recreate correlations between variables existing only in different files is dependent on some respective assumptions which are unlikely to be satisfied in practice. These theoretical properties are illustrated by results from a simulated match where the estimates of correlations, as well as other measures of association, are found to be grossly inaccurate.