A new algorithm, based on embedding phase space, to detect the P-wave characteristic points of an ECG signal is reported in this paper. The multi-lead ECG is transformed into points of an embedding phase space where similar ECG morphologies are converted into phase space points that are close using some distance measure. The algorithm is robust with respect to the type of selected characteristic points (onset, peak and end), morphology changes, baseline oscillations and high frequency noise. The performance of the algorithm has been successfully validated using both simulated and real ECG signals.