The purpose was to determine the accuracy of polyp measurement by endoscopy and CT. A colonic phantom was constructed containing 12 simulated polyps of known diameter. Polyp diameter was estimated during endoscopy by two observers independently. The phantom was then scanned using a 64-detector-row machine and diameter estimated by a further two observers independently, using 2D and 3D visualisation methods. All measurements were obtained twice. Bland-Altman statistics were used to assess agreement between observers' estimates and the reference diameter. The mean difference between observers' measurements and the reference diameter was smallest for estimates made using 3D CT (-0.09 mm and -0.03 mm) and greatest for endoscopy (-1.10 mm and -1.19 mm), with 2D CT intermediate. However, 95% limits of agreement were largest for 3D CT estimates (-4.38 mm to 4.20 mm). Estimates by 2D CT consistently overestimated polyp diameter, whereas endoscopy consistently underestimated diameter. In contrast, measurements by 3D CT were a combination of over- and under-estimates, with a tendency for disagreement to increase with the size of the polyp. The effect of observer experience was small and repeatability was best for 2D CT. Measurement error was encountered with all three modalities tested. Estimates made by 2D CT were believed to offer the best compromise overall.