The purpose of this study was to determine the optimal scanning technique for lesion detection in a small bowel phantom and to evaluate the virtual endoscopy (VE) technique in patients. A small bowel phantom with a fold thickness of 7 mm and length of 115 cm was prepared with nine round lesions (3 x 1 mm, 2 x 2 mm, 2 x 3 mm, 2 x 4 mm). Spiral CT parameters were 7/7/4, 3/5/2, 3/5/1, 1.5/3/1 (slice thickness/table feed/reconstruction interval). VE was done using volume rendering technique with 1 cm distance between images and 120 degrees viewing angle. Two masked readers were asked to determine the number and location of the lesions. Seven patients underwent an abdominal CT during one breathhold after placement of a duodenal tube and filling of the small bowel with methyl cellulose contrast solution. VE images were compared with the axial slices with respect to detectability of pathology. With the 7/7/4 protocol only the 4-mm lesions were visualised with fuzzy contours. The 3/5/2 protocol showed both 4-mm lesions, one 3-mm lesion and one false positive lesion. The 3/5/1 protocol showed both 4-mm and both 3-mm (one uncertain) lesions with improved sharpness, and no false positive lesions. One 2-mm and one 1-mm lesion were additionally seen with the 1.5/3/1 protocol. Path definition was difficult in sharp turns or kinks in the lumen. In all patients, no difference was found between VE and axial slices for bowel pathology; however, axial slices showed 'outside' information that was not included in VE. We conclude that the 3/5/2 protocol may be regarded as an optimal compromise between lesion detection, coverage during one breathhold, and number of reconstructed images in patients; round lesions of 4 mm in diameter can be detected with high certainty.