The output factor for a therapy photon beam is defined as the dose per monitor unit relative to the dose per monitor unit in a reference field. Convolution models for photon dose calculations yield the dose in units normalized to the incident energy fluence with phantom scatter intrinsically modeled. Output factors calculated with the convolution method as the dose per unit energy fluence relative to the calculated dose per unit energy fluence in a reference field could deviate as much as 5% if corrections are not made for perturbations due to treatment head scatter. Significant perturbations are particles backscattered from the collimators to the monitor and photons forward scattered from the filter and collimators in the treatment head. The forward scatter adds an "unmonitored" contribution to the total energy fluence of the beam. A model is developed that describes the field size dependence of these perturbations for conversion of output factors, calculated with the convolution method, to machine output factors as an integrated part in treatment planning. The necessary machine characteristics are derived from measurements of the output in air for a limited set of field sizes. The method has been tested using five different multileaf collimated irregular fields at 6 MV and for a large set of rectangular fields at 5, 6, and 18 MV and found to predict output factors with an accuracy better than 1%.