128 Magnetic resonance (MR) investigations of single or multifocal nodular liver lesions were retrospectively reviewed. All lesions had been identified, but not characterized, with ultrasonography (US). All the studies were performed with a 0.5-T superconductive magnet (Philips Gyroscan); spin-echo (SE) T1/proton density/T2-weighted and inversion recovery (IR) pulse sequences were used routinely. Characterization was attempted considering the following variables: a) lesion outline; b) the presence of some kind of capsular or pseudocapsular ring; c-d) homogeneity of signal intensity and its difference from surrounding liver parenchyma; e) possible central scar and its signal features; f) associated lesions (multifocal nodules, ascites, locoregional adenopathies, venous thrombosis). Diagnostic confirmation was obtained by means of biopsy (63 patients), of other imaging techniques (35 patients), or of clinical follow-up over 12 months at least (30 patients). Our results confirm high MR accuracy in the diagnosis of hemangioma (48/50 cases, 96% confidence) and even higher accuracy in focal fatty liver infiltration (9/9 cases, 100% confidence), thanks to some typical MR signal patterns on appropriate acquisition techniques--i.e., SE multiecho pulse sequences and IR sequences, respectively, with liver and fat signal nulling. Primary non-malignant focal liver lesions were identified mainly on a morphological basis (smooth roundish outline with/without capsular or pseudocapsular ring; central starlet scar; "basket" or "spoked wheel" patterns): these features allowed the correct identification of 5/7 focal nodular hyperplasia cases. On the other hand, in the absence of these typical morphological features and of specific MR signal changes, adenomas were misdiagnosed in all cases but one. The study of focal lesions in cirrhotic liver disease exhibited 66.6% confidence in the diagnosis of regenerating nodules, on the basis of their iso/hypointensity relative to liver on T2-weighted pulse sequences. Such a behavior seems to be due to intracellular iron loading, to small cell size and to thin vascular network, which are typical of cirrhotic regenerating areas. The diagnosis of hepatocellular carcinoma relies on both morphostructural features and possible associated lesions: in our series, 22/25 cases (88% confidence) were correctly identified. Indeed, this result was somehow influenced by the case history of the patients and by specific serologic indexes. Finally, MRI exhibited high sensitivity in the detection of focal liver involvement in neoplastic patients. However, the intrinsic range of variability and the lack of specificity of MR signal intensity, because of different histopathologic cell types, do not usually allow an unquestionable diagnosis to be made, especially for single lesions.