The aim of this study was to assess the accuracy of pelvic radiotherapy during a trial of blocked radiotherapy at the Royal Marsden Hospital, UK. Prospective evaluation was performed on 90 patients receiving CT planned pelvic radiotherapy using weekly anterior-posterior and lateral portal films. Field placement errors (FPEs) were calculated by comparing field centres of each film with a designated point of interest. Data was evaluated to calculate the overall treatment simulator differences, the number of error free treatments, and mean treatment-simulator position and to evaluate the role of systematic versus random errors. Age, weight, disease site, position of treatment, fractionation, blocked versus conventional techniques were assessed for their effect on treatment accuracy. The mean absolute error between treatment and simulator films was anterior right-left (ARL) 0.25 cm, anterior superior-inferior (ASI) 0.32 cm, lateral anterior-posterior (LAP) 0.42 cm, and lateral superior-inferior (LSI) 0.28 cm. On average the field centre was displaced by 0.66 cm (standard deviation, S.D. = 0.34) from that intended. On each treatment day 29% of anterior films and 45% of lateral films had at least one 0.5 cm error. Overall 59% of treatments had at least one 0.5 cm error and 9% a 1.0 cm error. The field centre was more than 0.5 cm from the position intended in 66% of treatments and over 1 cm for 14% of treatments. Analysis of variance showed that both random and systematic errors occurred in all directions. Though random errors were of similar magnitude in all direction (variance sigma 2 = 0.06-0.09 cm2); systematic errors showed a 4-fold variation being greatest in the LAP direction (sigma 2 = 0.19 cm2) and least the ARL direction (sigma 2 = 0.048 cm2). No factor consistently predicted for worse outcome in all directions. Hypofractionated treatments were less accurate in the LSI direction (P > 0.05). Systematic errors were associated in the ARL direction with hypofractionation (P < 0.01) and, in the LSI direction with weight (P < 0.03) and age (P < 0.05). We conclude that significant random and systematic errors can occur during pelvic radiotherapy especially in the LAP direction. These results suggest that in the absence of a customised immobilisation device, to cover 95% of errors, margins of 0.6 cm for RL and SI directions and 0.9 cm for AP direction should be allowed between the planning and clinical target volumes. However, ideally, each centre should determine their own margin requirements according to local clinical practice.