Different mechanical devices have been developed to improve cardiopulmonary resuscitation (CPR). The aim of this study was to evaluate active compression-decompression (ACD) CPR applied by Emergency Medical Service (EMS) in a defined population. The Trondheim region EMS (population 154,000) employs simultaneous paramedic and physician response. Upon decision to treat, patients with cardiac arrest of presumed cardiac origin were allocated to ACD CPR (CardioPump) or standard CPR by drawing a random number tag. Outcome in each patient was determined on a 5 point ordinal scale (no clinical improvement = 1, survival to discharge = 5). In 4 years, CPR was attempted in a total of 431 cardiac arrests, 54 patients (13%) survived to discharge; 302 patients with similar baseline characteristics were randomised. The prevalence of bystander CPR was 57% and the median call-arrival interval 9 min. By intention to treat, the mean score in the standard CPR group was 2.51 and 17/145 patients (12%) survived. The mean score in the ACD CPR group was 2.53 (P = 0.9) and 20/157 patients (13%) survived. Cerebral outcome was similar in the two groups. Among the 145 ACD patients, the technique was successfully applied in 110, found inapplicable in 35 and in seven patients chest compressions were unnecessary. This is the largest, single-centre, randomised, population based study of ACD CPR in out-of-hospital cardiac arrest to date. Even when considering a wider outcome spectrum than crude survival, we found no evidence of clinical benefit. In a quarter of cases ACD CPR was inapplicable, further limiting its potential usefulness.