Patient set-up in external beam radiotherapy to the thorax is routinely checked by matching anterior and lateral ports to simulator or digitally reconstructed radiograph (DRR) reference images. We report a method to derive bed shifts from matching multiple oblique treatment ports, exploiting data redundancy for match consistency checking. Portal images were acquired for 14 thorax patients on the first 3 treatment days and matched to DRRs. As per clinical practice, anterior and lateral portal images were matched and checked. In addition, treatment ports were acquired and matched and field placement errors calculated from equations generating lateral and ventrodorsal bed shifts from any pair of treatment fields. We compared bed moves obtained from clinical imaging and from oblique treatment fields. Inconsistencies larger than 3 mm triggered a further independent check using new reference anatomy. With clinical image matching, set-up errors had a 95% confidence interval of +/-8 mm ventrodorsal and +/-6 mm in the lateral direction. With independent oblique field re-evaluation, the confidence intervals were reduced to +/-3 mm and +/-2 mm, respectively. Matching using the oblique fields detected set-up errors greater than 5 mm in 16 out of 90 matches. In the clinical environment, match errors greater than 5 mm in the thorax area can remain undetected using anterior and lateral fields only. Consequently, necessary bed shifts are not made or made in error. We developed a technique that uses portal images from multiple oblique treatment fields and exploits data redundancy for internal match consistency checking.