The objectives of this study were to measure by in-vivo techniques the radiation doses received by the anorectum during pelvic radiotherapy and compare these with doses predicted by a GE TARGET™ treatment planning system. Nine patients with cancers of the prostate, bladder, cervix or uterus were planned with computed tomography (CT) using the TARGET system. A Scanditronix rectal probe containing five n-type photon-detecting diodes was placed in the anorectum during the planning CT scans. The probe position was standardized with the five diodes at 2 cm intervals from the anal verge. The probe diodes were calibrated for 10 MV photons. Doses were measured for each diode for two consecutive fractions in the first four patients and for five consecutive fractions in the remaining five. Thermoluminescent dosimeters were used initially to verify diode doses. The TARGET and diode measured doses were compared. In all patients diodes situated in the target volume were within 7% of predicted doses. This improved to 2.5% after measurement on five fractions. At the edges of the target volume, wide variability existed between measured and predicted doses (measured dose range -68% to +68% of predicted dose). Outside the target volume, considerable doses (up to 0.3 Gy per fraction) were measured in the anal canal, which were not predicted by TARGET. We conclude that TARGET planned doses are accurate within the confines of the target volume. The greatest variability was seen at the edges of the target volume, where dose can vary by 50% across a 1 cm distance in the anterior-posterior plane. TARGET does not account for scattered dose beyond the field edges and therefore underestimates the dose received by the anal canal.