Separation of the effects of extraneous variables from the effects of a factor under study (often termed control of confounding) is one of the key prerequisites for validly estimating the magnitude of the study factor's effects. Because of the phenomenon of confounding by indication, confounding of effects of different factors is a common problem in the assessment of medical technology. We give several examples illustrating that the decision of whether a recorded variable is a confounder in a data-set must be decided on the basis of subject-matter knowledge and clinical judgement. There is no alternative to use of such judgement; statistical selection procedures based on significant tests, such as stepwise regression, can be particularly misleading.