To evaluate the sensitivity of sonography, CT, and MR imaging in the detection of hepatic masses in carcinoma patients, we conducted a prospective study of 75 consecutive patients with gastrointestinal tumors who were admitted for surgical resection of the primary tumor. Sonography was performed with convex transducers of 3.5 and 5.0 MHz. Three noninvasive CT techniques were used: unenhanced CT scans, the incremental bolus dynamic scanning technique, and delayed scanning 4-6 hr after bolus injection of 60 g of iodine. MR images (1.5 T) were acquired as presaturated T1- and T2-weighted spin-echo sequences and as breath-holding fast low-angle shot (FLASH) 60 degrees and FLASH 15 degrees sequences. As it is difficult to distinguish benign from malignant masses solely on the basis of morphologic criteria, the techniques for each imaging method were designed to detect and not to characterize hepatic lesions. Each examination was interpreted blindly, and the results were compared with surgical findings, intraoperative sonography, and biopsy of the liver as the gold standard. All focal hepatic masses verified at surgery, malignant or benign, were included in the analysis. Sixty-five (68%) of 95 focal hepatic masses were detected by CT, 60 lesions (63%) by MR, and 50 lesions (53%) by sonography. Although lesions 1-2 cm were shown almost equally well by CT and MR (74% and 77%, respectively), the detection rate of smaller lesions (less than 1.0 cm) decreased more drastically with MR (31%) than with CT (49%). Sonography had a sensitivity of only 20% with the smaller lesions. All imaging techniques had a sensitivity of 100% for focal hepatic masses larger than 2.0 cm. Our results show that CT has a higher overall sensitivity (68%) than MR and sonography for the detection of focal hepatic masses. When the results of the three procedures are combined, the overall sensitivity is 77%. This is unsatisfactorily low, as CT and MR have a size threshold of about 1.0 cm and are relatively unreliable for the detection of smaller lesions.