Overdiagnosis is an emerging problem in health policy and practice: we address its definition and ethical implications. We argue that the definition of overdiagnosis should be expressed at the level of populations. Consider a condition prevalent in a population, customarily labelled with diagnosis A. We propose that overdiagnosis is occurring in respect of that condition in that population when (1) the condition is being identified and labelled with diagnosis A in that population (consequent interventions may also be offered); (2) this identification and labelling would be accepted as correct in a relevant professional community; but (3) the resulting label and/or intervention carries an unfavourable balance between benefits and harms. We identify challenges in determining and weighting relevant harms, then propose three central ethical considerations in overdiagnosis: the extent of harm done, whether harm is avoidable and whether the primary goal of the actor/s concerned is to benefit themselves or the patient, citizen or society. This distinguishes predatory (avoidable, self-benefiting), misdirected (avoidable, other-benefiting) and tragic (unavoidable, other-benefiting) overdiagnosis; the degree of harm moderates the justifiability of each type. We end with four normative challenges: (1) methods for adjudicating between professional standards and identifying relevant harms and benefits should be procedurally just; (2) individuals, organisations and states are differently responsible for addressing overdiagnosis; (3) overdiagnosis is a matter for distributive justice: the burdens of both overdiagnosis and its prevention could fall on the least-well-off; and (4) communicating about overdiagnosis risks harming those unaware that they may have been overdiagnosed. These challenges will need to be addressed as the field develops.