The standard ECG remains the most common non-invasive tool for assessing atrial fibrillation. Specific signal processing techniques have been developed to improve the diagnosis. However, validation of such tools is challenging and comprehensive invasive data may not easily be obtained. To facilitate this task, we developed a computer model of the atria. In this electrophysiological model, atrial fibrillation was simulated and the manifestation of its electrical activity on the thorax was computed. The resulting realistic-looking synthetic ECG signals were used as benchmarks for testing, evaluating and comparing ECG processing techniques such as cancella-tion of the ventricular activity, vectorcardiography and domi-nant frequency analysis.