This paper describes a technique for the matching of population samples. The matching is achieved through prediction and the method turns-upon the combining of empirically developed predictors to give the best available predictive (or matching) composite. The development of this composite is through the principle used in-simple biological classification, which is to be sharply distinguished from the principle inherent in multiple correlation. The mathematical bases of the method are very simple and appear to be superior to the usual maximising formulas. Empirical matching is to be contrasted with matching by custom or hunch.