Despite current selection criteria (NYHA Class III-IV, LVEF < 35%, QRS > 120 ms with LBBB), 30% of patients do not benefit from cardiac resynchronization therapy (CRT). The use of QRS duration as selection criteria for CRT has not been evaluated systematically yet. Accordingly, the value of QRS duration at baseline (and reduction in QRS duration after CRT) to predict responders was studied. Patients were evaluated at baseline and after 6 months of CRT for NYHA Class, quality of life score, and 6-minute walk test. QRS duration was evaluated before, directly after implantation, and after 6 months of CRT. Sixty-one patients were included; 45 (74%) patients were classified as responders (improvement of NYHA Class, 6-minute walking distance and quality of life score) and 16 (26%) as nonresponders. QRS duration at baseline was similar between the two groups: 179 +/- 30 ms versus 171 +/- 32 ms, NS. Directly after implantation, QRS duration was reduced from 179 +/- 30 ms to 150 +/- 26 ms (P < 0.01) in responders; nonresponders did not exhibit this reduction (171 +/- 32 ms vs 160 +/- 26 ms, NS). After 6 months of CRT, QRS shortening was only observed in responders (from 179 +/- 30 ms to 159 +/- 25 ms, P < 0.01). ROC curve analysis showed that a reduction in QRS duration > 10 ms had a high sensitivity (73%) with low specificity (44%); conversely, a > 50 ms reduction in QRS duration was highly specific (88%) but not sensitive (18%) to predict response to CRT. No optimal cutoff value could be defined. QRS duration at baseline is not predictive for response to CRT; responders exhibit a significant reduction in QRS duration after CRT, but individual response varies highly, not allowing adequate selection of responders.