Purpose: To demonstrate that in clinical diagnosis with more than two candidate diseases, the specificity of a sign or symptom for a particular disease must change as other findings are observed, and its sensitivity is likely to change. Method: Theoretical argument concerning Bayes' theorem, demonstrated using probability equations. Results: The simple account of the impact that a medical test result has upon the probability a patient has a disease talks of the sensitivity p(positive test|disease) and specificity p(negative test|no disease) of the test as universal, stable quantities. When we describe differential diagnosis for a clinical presentation as the use of available signs and symptoms to decide which of many diseases a patient has, however, the situation compels specificity to be a changing quantity. The specificity of a finding for a particular diagnosis is constituted of its sensitivities for the other diagnoses, weighted by their probabilities. If the relative probabilities of the other diseases change, and they have different sensitivities, then the target disease's specificity must change. Further, if the set of signs and symptoms related to a particular disease are interdependent (Fryback, 1978), then the sensitivity of a sign, for a disease, can change as other information is gathered about the patient. This in turn can be a second reason the specificities change. Conclusions. Hence the specificity and probably the sensitivity of a clinical sign or symptom can change dynamically as information about other symptoms is gathered. This undermines the assumptions of the 2 by 2 table and the odds-likelihood forms of Bayes' theorem, of ROC analysis, and of SPin and SNout rules, as applied to diagnosis using clinical signs. It does not affect the validity the assumptions of clinical prediction rules based on multivariate analyses of clinical data sets. Research investigating the magnitude of the changes for particular diagnoses is needed, to determine when this actually matters.