Ultrasound is a non-invasive image modality which allows for real time acquisition. Nevertheless, the low quality of the acquired images makes this a difficult-to-interpret modality during surgical procedures. To overcome this, the registration of ultrasound images with preoperative CT or MR images has been routinely used to fuse complementary information. This work presents the evaluation of eight similarity measures used in the registration of ultrasound and CT images of the left atrium and the pulmonary veins. Each intensity-based similarity measure was evaluated computing its accuracy, capture range, distinctiveness of the optimum, risk and non-convergence and number of minima. The results show that the Woods criterion presents a globally better performance than the other similarity measures. This is especially true for the accuracy and distinctiveness of the optimum indicators. Preprocessing US images does not improve the performance of all similarity measures, except for Woods criterion that shows the optimal accuracy.