The paper approaches criterion-referencing as a problem of validating tests for particular inferences about human behavior. Some definitions of the term "criterion-referenced" are reviewed, and the position is taken that direct inferences about what a test-taker can or cannot do--criterion-referenced inferences, that is--need not be restricted to tests that are composed of actual samples of the behaviors of interest. Primary attention is given to the criterion of minimal competency in some significant educational area, but other applications are also discussed. Several methods are suggested for validating tests for making inferences to a particular criterion or to several criteria of interest. Some of the limitations of the methods are discussed, and the suggestion is made that more than one method be used to validate any desired criterion-referenced inference. (Author)