Two cubic-shaped phantoms of water equivalent (WE) material, one homogeneous and one with a lung substitute were used to simulate an intact breast. They were irradiated with a constant dose using an isocentric tangential field of 6- and 18-MV photons, respectively. The absorbed dose was measured at the isocentre for a range of the lateral distances of the isocentre from the edge of the phantoms. Four currently available treatment planning systems (TPS), two with a 2-dimensional (2-D) and two with a 3-dimensional (3-D) algorithm were used to calculate the dose at same points in each phantom. A comparison of the results showed that for the homogeneous phantom, the 2-D algorithms over-estimated the dose by up to 10% for 6-MV photons at an isocentre depth of 1 cm laterally below the surface and 3.6% for 18-MV photons at 2 cm below the surface. For the 3-D algorithms, agreement with measurement was within +/-3% at all lateral isocentre depths for both energies. For the inhomogeneous phantom, as expected, the differences were generally greater as all 4 TPS ignore electron transport and photon scatter from heterogeneities. Agreement with measurement generally improved with increasing lateral depth of the isocentre below the surface. Calculations using an anatomical breast phantom showed that when changing from a 2-D to a 3-D algorithm, differences of 5-10% between the prescribed dose and the dose delivered to the patient can be expected, depending on the algorithm used, the photon energy and the lateral depth of the dose reference point below the skin.