PURPOSE: To assess, by means of CT-pathologic correlation, the ability of CT to detect hepatic VX-2 carcinomas in rabbits, and to determine the factors influencing the sensitivity of tumor detection. MATERIALS AND METHODS: By means of direct portal inoculation, VX-2 carcinomas were implanted in the liver of eight rabbits after laparotomy. Two weeks later, dual-phase spiral CT scanning was performed using scan parameters of 3-mm collimation and reconstruction intervals of 1 mm and 3 mm. Radiologic-pathologic correlation involved the comparison of CT images and pathologic slices. The sensitivity and positive predictive value with which each technique detected the presence of tumors were calculated. RESULTS: Using pathologic slices, 2-40 (mean, 9.1) mm in size, 65 tumor nodules were detected. Overall sensitivity and the positive predictive value were 63% and 73%, respectively. Sensitivities for tumors of 2-5 mm and 6-10 mm were 28% and 79%, respectively. For images reconstructed with a 1-mm interval, sensitivity was significantly higher than that where a 3-mm interval was used (79% vs. 46%), but sensitivities for arterial and portal-phase imaging were not significantly different. Among small tumors (< or = 10 mm), 63% (12/19) showed strong enhancement at arterial-phase imaging. Regarding tumor detection, inter-observer concordance between the three radiologists was excellent for portal-phase images (k=0.86 and 0.83 for 1 mm and 3 mm reconstruction intervals, respectively), and good for arterial-phase images (k=0.77 and 0.73 for 1 mm and 3 mm reconstruction intervals, respectively. CONCLUSION: Even where dual-phase scanning with 3-mm collimation is used, spiral CT is limited in its ability to detect tumors 5 mm or less in diameter, though overlapping reconstruction improves the sensitivity with which those of 6-10 mm are detected. In the evaluation of arterial-phase images obtained at contrast-enhanced dual-phase imaging, special attention should be given to small hyper-attenuating nodules.