Although IMRT has been shown clinically to increase skin doses for some patients, it has also been shown that intensity modulated delivery does not, of itself, increase skin doses. The reason for this apparent difference is that inverse planning can result in solutions that give high fluence to tangential beam segments near the skin surface, in an attempt to counter the build-up region. In cases where the clinical target volume (CTV) stops short of the skin surface, but the planning target volume (PTV) does not, there is no clinical reason to treat the skin. The CTV-PTV margin exists purely to ensure that fields are large enough to allow for geometrical uncertainties. With an objective function based on the doses to the PTV, it is possible for a plan that gives excess fluence to the skin to have a lower objective function, and hence to be preferred in an optimization. We describe I technique of plan evaluation, based on analysis of a plan by recalculating several plans in which the isocentre has been offset by a distance equal to the CTV-PTV margin. We demonstrate that changes to a plan that reduce a PTV-based objective can give a worse dose distribution to the CTV when systematic and random set-up errors are accounted for, and increase skin dose. Several possible strategies for avoiding this problem are discussed, including the use of the skin as an organ at risk, modification of the PTV to avoid the skin, and the use of 'pretend bolus' applied in planning but not in treatment. The latter gave the best results. The possibility of using the evaluation method itself, as the basis of an objective function for optimization, is discussed.