Objective: To assess the validity of early sensory evoked potential (SEP) recording for reliable outcome prediction in comatose cardiac arrest survivors within 48 h after restoration of spontaneous circulation (ROSC). Design and setting: Prospective cohort study in a medical intensive care unit of a university hospital. Patients: Twenty-five comatose, mechanically ventilated patients following cardiopulmonary resuscitation Measurements and results: Median nerve short- and long-latency SEP were recorded 4, 12, 24, and 48 h after ROSC. Cortical N20 peak latency and cervicomedullary conduction time decreased (improved) significantly between 4, 12, and 24 h after resuscitation in 22 of the enrolled patients. There was no further change in short-latency SEP at 48 h. The cortical N70 peak was initially detectable in seven patients. The number of patients with increased N70 peak increased to 11 at 12 h and 14 at 24 h; there was no further change at 48 h. Specificity of the N70 peak latency (critical cutoff 130 ms) increased from 0.43 at 4 h to 1.0 at 24 h after ROSC. Sensitivity decreased from 1.0 at 4 h to 0.83 at 24 h after ROSC. Conclusion: Within 24 h after ROSC there was a significant improvement in SEP. Therefore we recommend allowing a period of at least 24 h after cardiopulmonary resuscitation for obtaining a reliable prognosis based on SEP.