Capnography is a valuable tool in the management of cardiac arrest, since end-tidal CO2 (PetCO2) correlates well with cardiac output and there are no other suitable noninvasive ways to measure this important variable during resuscitation. Animal studies also suggest that PetCO2 correlates well with the likelihood of resuscitation, but this has never been confirmed in humans. We prospectively studied 55 adult, nontraumatic prehospital cardiac arrest patients. PetCO2 was monitored with an in-line sensor on arrival in the ED and throughout the arrest, which was managed by the usual advanced cardiac life-support treatment guidelines. Chest compression was carried out mechanically. Patients were assessed for return of spontaneous pulse as evidence of initial resuscitation; hospital discharge and long-term survival were not examined. Fourteen patients developed spontaneous pulses and were resuscitated, and 41 were not. The length and aggressiveness of treatment and CPR were not different between the two groups, nor were there differences in down time, resuscitation time, or other factors known to affect outcome. Patients who developed a pulse had a mean PetCO2 of 19 +/- 14 (SD) torr at the start of resuscitation, and those who did not had a mean PetCO2 of 5 +/- 4 torr (p less than .0001). This difference was significant both in nonperfusing rhythms (asystole and ventricular fibrillation) and in potentially perfusing rhythms (electromechanical dissociation). An initial PetCO2 of 15 torr correctly predicted eventual return of pulse with a sensitivity of 71%, a specificity of 98%, a positive predictive value of 91%, and a negative predictive value of 91%. A receiver operating curve was generated for sensitivity and specificity of the test at varying PetCO2 thresholds.