Objective. To determine whether the combination of prolonged QRS duration and brain natriuretic peptide (BNP) levels predicts left ventricular systolic dysfunction (LVSD) with a higher accuracy compared with QRS duration or BNP alone. Setting. University hospital. Methods. We studied 128 consecutive patients with suspected cardiac disease. At rest the QRS duration on 12-lead ECG and BNP levels were determined. A left ventricular ejection fraction (LVEF) <50% at echocardiography was defined as LVSD. Results. QRS duration in the LVSD group (n ¼ 66, LVEF 30 ± 8%) was longer than in the group without LVSD (n ¼ 62, LVEF 60 ± 5%; QRS 129 ± 34 vs. 96 ± 20 ms, P < 0.001). BNP was higher in the LVSD group compared with controls (467 ± 397 vs. 169 ± 242 pg mL )1 , P < 0.001).A QRS duration of >0.1, >0.11 or >0.12 s was highly specific (63, 90 and 98%) but less sensitive (84, 81 and 75%) for the prediction of LVSD. A QRS cut-off value of 106 ms was moderately sensitive (65%) but very specific (87%) for the prediction of LVSD, whereas a BNP cut-off value of >84 pg mL )1 was highly sensitive (89%) but only modestly specific (58%). The positive likelihood ratio for LVSD of abnormal BNP (2.0) and QRS prolongation >0.1 s (2.3) was improved by the combination of both criteria (5.1). In multivariate analysis, BNP and QRS duration were independent predictors of LVSD.Conclusions. The combination of abnormal BNP and QRS prolongation yields a higher positive likelihood ratio for the detection of LVSD compared with the two criteria alone.