During the Heartstart Scotland project all 407 ambulances in Scotland were equipped with automated external defibrillators (AEDs). All cases of chest pain or collapse aged over 10 years were monitored and multiple 3-s rhythm strips recorded in a solid state memory module. A shockable rhythm was defined as an organised rhythm of > or = 180 beats/min or a disorganised rhythm of > or = 100 beats/min and amplitude > 0.1 mV. We analysed all the stored rhythm strips in two patient populations to determine the ability of the AED and ambulance crews to detect a shockable rhythm and to initiate appropriate defibrillation. The first population comprised 493 patients, all of whom had received shocks. A total of 4741 rhythm strips were analysed, of which 1461 were true positives, 33 false positives, 3161 true negatives and 86 false negatives. Overall sensitivity of the AED was 94.4% and specificity 99.0%. The second population comprised a random sample of 200 shocked and 200 non-shocked arrests. The combined group contained 4154 rhythm strips of which 562 were true positives, 12 false positives, 3460 true negatives and 120 false negatives. Overall sensitivity of the system (AED+crew) was 82.4% and specificity 99.7%. However, only 66 of the 120 false negatives were attributable to the AED giving a sensitivity of 90.3% for the AED. The sensitivity of the AED is dependent on the prevalence of shockable rhythms, but will be within the range 90.3-94.4% for most emergency medical services. We conclude that early management of potentially lethal arrhythmias by ambulance technicians using AEDs is practical with acceptable sensitivity and specificity.