Three-dimensional ultrasound localization has been performed for external beam prostate treatments at our institution since September 2001. This article presents data from the daily shifts for 221 patients and 5005 fractions, and the results of tests performed to assess the system's performance under clinical conditions. Three tests are presented: (1) To measure the accuracy of the shifts, eight patients treated on a helical tomotherapy machine were localized daily using both ultrasound (US) and a megavoltage computed tomography (MVCT) scan. Comparison of the shifts showed that US localization improved alignment for six of the eight patients when compared to alignment using skin marks alone. The mean US-MVCT vector for these six patients was 3.1+/-1.3 mm, compared to 5.1+/-2.1 mm between the MVCT and the skin marks. The other two patients were identified as poor candidates for US prior to their first treatment fraction. (2) To assess the extent of intrafraction motion, US localization was repeated after treatment for six patients and a total of 29 fractions. The mean intrafraction prostate shift was 1.9+/-1.0 mm, and the shift was within the 3 mm localization uncertainty [Tomé et al., Med. Phys. 29, 1781-1788 (2002); in New Technologies in Radiation Oncology, edited by W. Schlegel, T. Bortfelde, and A. Grosu (Springer, Berlin, 2005)] of the system for 25 of 29 fractions. (3) To assess the interuser variation in shifts, four experienced operators independently localized five patients for five consecutive fractions. The standard deviation of the users' shifts was found to be approximately the same as the system's localization uncertainty. For shifts larger than the system localization uncertainty, the standard deviation of the users' shifts was nearly always much smaller than the mean shift. Taken together with the results of the US-MVCT comparison, this indicates that the shifts improved patient localization despite differences between users.