Two approaches can be taken to improve the performance of an automatic arrhythmia detector: perfecting the detection algorithms or improving the quality of the investigated traces by preprocessing the original traces. This paper reports on the results of a data preprocessing approach. Preprocessing consists in constructing new traces, which we call virtual. They are mathematically obtained from the original traces and referred to the dominant cardiac electric axis. The classifications obtained with an arrhythmia detector using both virtual and original traces are presented and discussed. By comparing the performance indices obtained under the two different conditions, it can be seen that a diagnosis based on the virtual traces is as acceptable as one based on the original traces. This result should be judged as favorable, since the algorithm was not adjusted or calibrated to the virtual traces, while those who developed it had certainly calibrated the parameters to the original traces.