This paper identifies the kinds of behaviour changes which are commonly included under the heading of "learning". It argues that the term refers to the behaviour changes of individuals and that behaviour can change along any of several dimensions. The paper also identifies the main technical requirements which need to be met if learning is to be measured in a scientific manner. These include measurement accuracy, measurement sensitivity and the use of standard units to report measurement results. When learning is tested only once or twice, these requirements are almost never met. The case for continuous measures of learning is illustrated using a variety of student-administered learning experiments. The investigator who wishes to identify variables which affect student learning must first develop a measure of learning. In the great majority of learning experiments conducted this century, learning has been measured by developing a test of learning outcomes which is then administered either once (at the end of instruction) or twice (at the beginning and at the end of instruction). This paper agues that this procedure may be adequate to the task of measuring achievement but that it is an extremely unsatisfactory way of measuring learning.