It is recognized at the outset that since most of the investigations which social scientists carry out are multivariate by nature these scientists should have a sound working knowledge of the more commonly required multivariate techniques. In acquiring this, some understanding of matrix algebra is essential and suggestions are made about how the subject may be taught. Attention is then given to the imprecise nature of most of the measuring instruments available to social scientists and to the need for models which make explicit provision for errors of measurement in their data. Some commonly used models are mentioned and the pitfalls inherent in their uncritical use are illustrated by examples. The importance of validating results either by replication or by the use of parallel samples is emphasized, but it is shown that these do not provide a panacea for all ills. Finally, the advantages of using transformations when coping with multiway contingency tables are mentioned.