To evaluate repeat radiosurgery (RS2) for cerebral arteriovenous malformations (AVMs) after failure of initial radiosurgery (RS1). Between 1986 and 2000, 41 patients underwent RS2. Nine patients were not assessable: seven had insufficient follow-up (RS2 in 1999 and 2000) and two had no recent control angiography data. Thus, 32 (78%) of 41 patients were assessed. Most lesions (29 [90%] of 32) were supratentorial: 22 (69%) on the left, 8 (25%) on the right, and 2 on the midline (6%). The patients had Spetzler-Martin Grade 1-5 (median Grade 3). The symptoms before RS1 included hemorrhage in 20 (63%), epilepsy in 10 (31%), progressive neurologic deficits in 2 (6%), and headaches in 6 (19%). Five patients had two or more symptoms. Twenty-two patients (69%) had received other treatment before RS1, including neurosurgery in 3 patients (9%) and one to six embolizations in 19 patients. At RS1, the median largest nidus dimension was 2.7 cm (range 0.8-5). The median volume was 2.7 cm(3) (range 1.2-9.9). The median time from RS1 to RS2 was 52 months (range 12-126). Between RS1 and RS2, 7 (22%) of 32 patients experienced bleeding. The same irradiation technique was used for RS1 and RS2, except for 2 patients who underwent RS2 at another institution. Circular 15-MV X-ray minibeams (range 6-20 mm) and coronal arcs were used. RS1 was monoisocentric in 75% of cases and multiisocentric in 25%. At RS2, the median largest nidus dimension was 3 cm (range 1.4-5). The median volume was 4.2 cm(3) (range 0.8-13.4). RS2 was monocentric in 72% of cases and multiisocentric in 28%. After RS2, the median follow-up was 19.5 months (range 0-79; mean 25.3). After RS2, the obliteration rate was 59.3% (19 of 32). The median time to arteriographic obliteration was 21 months (range 12-96). The survival rate was 97% (31 of 32). Five of the 13 patients with a nonobliterated nidus experienced complications; 3 had bleeding (9%) and 2 without prior neurologic deficits developed partially regressive neurologic deficits. One patient with a previously existing deficit developed an additional new partially regressive neurologic deficit after an episode of bleeding. Thus, 3 (9%) of 32 patients had neurologic complications. Moderate-grade parenchymal changes at MRI increased after RS2 (88.2% vs. 57.7% after RS1; p = 0.10, not significant). However, necrosis-like changes did not significantly increase. After RS1 failure, salvage may be attempted by embolization, neurosurgery, or RS2. RS2 should be considered after the second successive annual angiogram if reduction of the nidus is <25%. The results after RS2 are encouraging. A multidisciplinary approach is mandatory to reduce the initial failure rate and to choose the modality and timing of salvage treatment.