Clinicians are increasingly facing complex diagnostic problems in patients who have a pre-existing disease and in whom new symptoms develop which may be complications of that disease or manifestations of another, perhaps more common, condition. The neurological complications of malignancy fall into this situation. The results of diagnostic tests applied under these circumstances run the risk of being misinterpreted, largely due to lack of appreciation of the powerful effect of the pre-existing disease on a priori estimates of disease which enter the differential diagnosis. A bayesian analysis outlines the basic structure of these novel problems--a transient neurological disturbance in a man with lung cancer but a negative computerised tomographic scan illustrating the points. Bayesian probability revision yields accurate estimates of the predictive values of tests that tend to be different from those generated by intuitive clinical reasoning.