There have been 2 major treatment approaches for patients with severe subarachnoid hemorrhage (SAH): aggressive treatment for all patients irrespective of grade, and treatment of patients selected based on certain indices. We treated 34 patients with poor-grade SAH classified by the Hunt and Kosnik criteria according to an aggressive protocol we had devised (A). To evaluate its efficacy, we compared the results with those obtained in 103 patients treated by the previous protocol (B) involving a selective strategy, and focused on the problems of treating patients with poor-grade SAH. Protocol A: Patients showing no flow in the initial angiogram received supportive care alone; the remaining patients underwent clipping surgery with or without external decompression, or coil embolization; hypothermia or normothermia (for elderly patients or those with complications) was introduced when intracranial pressure (ICP) exceeded 25 mmHg. Protocol B: Patients with SAH+hematoma causing a mass effect underwent emergency surgery; the remaining patients were pretreated for 12 h with control of blood pressure and intracranial pressure, and then radical surgery was performed on all patients who were Grade III or better and on patients at Grade IV below 75 years of age and without systemic complications. Mean arrival time of patients treated by protocol A was 4 h, compared with 1 h in the protocol B group, and the difference was significant. The proportion of patients who underwent radical treatment was 87.5% in A and 53.4% in B, and the difference was also significant. However, a favorable outcome classified as GR or MD by the GOS was seen in 43.8% of patients in group A and 42.7% of those in group B; the death rate was 28.1% and 36.9%, respectively, indicating no significant difference of outcome between the 2 protocols. Hypothermia was effective for management of ICP, but this carries a risk of vasospasm. SjO2 monitoring was useful for control of rewarming and also for predicting outcome during the ultra-early phase. The significantly delayed arrival of patients treated by protocol A suggests that a proportion of patients whose neurological grade may improve within several hours might be initially excluded. If so, the figure of 43.8% for patients achieving GR+MD in protocol A might be improved further.