Eighteen cases of rectal carcinoma were staged preoperatively with transrectal endosonography (EUS), CT and MRI (0.5 T). The results were compared with surgical specimens and histology to evaluate the accuracy of the imaging modalities in staging rectal carcinomas which had been quantified according to Astler-Coller's classification. All methods identified the lesion (100% sensitivity). EUS and MRI correctly staged 8 cases (44%) and CT 9 cases (50%). CT and MRI mistakes were relative to overstaging, whereas EUS understaged 4 cases (22%) and overstaged 6 cases (33%). In local tumor staging ("T" variable), CT and MRI understaged no lesions, thus exhibiting 100% sensitivity, which was higher than EUS sensitivity (92%). Conversely, CT and MRI more frequently overstaged the lesions, thus demonstrating lower sensitivity than EUS (55% and 50%, respectively, versus 76% for EUS). As for the "N" variable, EUS identified node metastases in one case only (25%) and misdiagnosed as positive 4 cases of negative node involvement. All the C-stage lesions were correctly diagnosed by CT and MRI (whose findings were in agreement) which also overstaged as C three cases with hyperplastic node enlargement. The diagnostic accuracy of EUS, which was highest for the A and B1 stages, progressively decreased for bigger lesions, clearly understaging node involvement. On the contrary, CT and MRI accuracy rates were lower in small tumors involving the rectal wall only, whereas they always identified tumor spread beyond the bowel wall into perirectal fat, and node metastases. Therefore, to conclude, EUS can be used first: in case of extraluminal tumor spread, CT is the method of choice, more accurate than MRI in identifying node involvement and equally effective in evaluating perirectal fat infiltration and pelvic structures involvement. Whenever the pelvic floor is involved, MRI is the best imaging method, thanks to its multiplanar capabilities, for better detailing of musculoskeletal involvement.