Recently, concerns have been raised about thephenomenon of 'overdiagnosis', the diagnosis of a conditionthat is not causing harm, and will not come to causeharm. Along with practical, ethical, and scientific questions,overdiagnosis raises questions about our concept ofdisease. In this paper, we analyse overdiagnosis as anepistemic problem and show how it challenges manyexisting accounts of disease. In particular, it raises questionsabout conceptual links drawn between disease anddysfunction, harm, and risk. We argue that 'disease' shouldbe considered a vague concept with a non-classical structure.On this view, overdiagnosed cases are 'borderline'cases of disease, falling in the zone between cases that areclearly disease, and cases that are clearly not disease. Wethen develop a preÂ´cising definition of disease designed toprovide practical help in preventing and limiting overdiagnosis.We argue that for this purpose, we can definedisease as dysfunction that has a significant risk of causingsevere harm to the patient.