A Monte Carlo code (NCF) has been developed to calculate the neutron vector flux emerging from a collimator of given composition and geometry and is applied here to the collimator of a Marconi Avionics Hiletron (d, t) facility using up to lo7 neutron histories. For variations in the collimator geometry and in the polythene, tungsten and boron carbide content, the calculations verified that the original Hiletron collimator was well optimised for high-energy neutrons in the field, that the neutrons backscattered into the field from the surrounding shielding are significant only in the 0-0.5 MeV energy group and that the induced activation of a collimator differentially loaded with tungsten will allow its use with a source intensity of l O I 3 S".A second code (MCCND) was used to calculate the dose deposited in a tissue-equivalent phantom using as input the neutron vector flux output from the collimator code NCF. A collimated (d, t) beam of field size 7 cm x 7 cm and an SSD of 0.8 m was assumed and up to lo7 neutron histories were followed. Dose in the phantom was binned for every 0.01 m increment in depth, except near the surface where 0.002m was used. Radial distances outwards from the beam axis were chosen such that there were three increments within the field, one in the penumbra and three outside the penumbra. Further binning of dose took place within 19 LET intervals between 0 and 1250 keV pm". Calculations were made of the neutron and gamma dose in the field (0-0.038 m) down the beam axis, the dose profile at the surface, the dose : LET distribution along and across the beam, the dose-and track-averaged LET values and the absolute kerma and dose rates. Our calculations predict that a rather larger fraction of total dose comes from heavy recoil events than is suggested by some, but not all, of the previous calculations. We believe that previous experimental measurements have also tended to underestimate the high LET contributions.0 -2 ~ while the distribution in brightness across the source disc was accounted for by choosing the point of emission randomly between 0 and 0.02 m from the centre to give a l / r distribution. An alternative Gaussian distribution was also used, although no significant change in emergent flux was observed.