AIMS: In CARE-HF, cardiac resynchronization therapy (CRT) lowered morbidity and mortality in patients with moderate to severe heart failure. We examined whether baseline and follow-up electrocardiographic characteristics might predict long-term outcome. METHODS AND RESULTS: CARE-HF randomly assigned 409 patients to medical therapy (MT) plus CRT, and 404 patients to MT alone. Electrocardiographic measurements were made at baseline during sinus rhythm, and at 3 months during paced or spontaneous rhythm depending on treatment assignment. Favourable outcome was defined as freedom from death, urgent transplantation, or cardiovascular hospitalization. Among patients assigned to CRT, 39% had unfavourable outcomes including 55 deaths. By single variable analysis, (i) prolonged PR interval, left QRS axis (but not QRS duration), and left bundle branch block (BBB) at baseline, and (ii) heart rate, PR, and QRS duration at 3 months predicted unfavourable outcome. By multiple variable analysis, treatment assignment (P = 0.0001), PR (P = 0.0004), and right BBB (P < 0.00013) at baseline predicted outcome, whereas baseline JTc and QRS duration at 3 months predicted all-cause mortality and heart failure hospitalization (P = 0.0071). CONCLUSION: In CARE-HF, QRS duration at baseline did not predict outcome, but QRS at 3 months was a predictor by single variable analysis. Patients with prolonged PR interval and the 5% of patients with right BBB had a particularly high event rate.