The problem of disease definition is related to the problem of proving that a certain agent is the necessary cause of a certain disease. Natural kind terms like ‘rheumatoid arthritis’ and ‘AIDS’ refer to essences which are discoverable rather than predeterminate. No statement about such diseases is a priori necessarily true. Because theories on necessary causes involve natural kind semantics, Koch's postulates cannot be used to falsify or verify such theories. Instead of proving that agent A is the necessary cause of disease D, we include A in a theoretical definition of D, take this to represent the real meaning of ‘D’, and discard the pretheoretical definition. This is illustrated by Koch's own attempt to prove he had discovered the necessary cause of tuberculosis. Methodological arguments about disease causation require a clear view of our use of diagnostic terms. Medical lexicographers should do more to provide such a view.