BACKGROUND AND PURPOSE To implement an on-line correction scheme based on implanted markers to reduce treatment margins in external beam radiation therapy (EBRT) of carcinoma of the prostate. In turn reduction in treatment margins reduces irradiated volumes and offers the possibility of reduced normal tissue complications or escalated target dose. PATIENTS AND METHODS Five or six gold markers were implanted in 10 patients treated for prostate carcinoma using EBRT. All patients were enlisted in an IRB-approved protocol. Before each fraction two portal images were obtained using a low dose (2MU). Positions of the markers were calculated from these images using an in-house developed program. Corrections were applied with a threshold of 2mm displacement. After correction the procedure was repeated. RESULTS Overall systematic errors were reduced from 7.45, 1.29, and 5.12 mm to 0.65, 0.11, and 0.46 mm in, respectively, the antero-posterior, lateral, and cranio-caudal directions. Likewise, the overall SD were reduced from 5.99, 5.34, and 4.44 mm to 2.82, 2.64, and 2.22 mm, respectively. All reductions were highly significant (P < 0.01) using a t-test for systematic and an F-test for random errors. On an individual level all but three patients showed significant improvements in all directions for the random errors. All patients improved in at least one direction. Systematic errors were significantly lower in all patients. Simulated correction schemes using this data suggest that margin reduction using off-line reduction does not benefit substantially from on-line corrections in the first few fractions. CONCLUSIONS Use of marker-based correction improves the patient position. Factors influencing the accuracy were: (1) number of seeds usable for correction, (2) distribution of markers throughout the volume of interest, and (3) objective instructions for patient realignment.