The objective of this paper is to evaluate the robustness of biometric approaches based on human ECG signals. Two questions are addressed: is it required to record the ECG in supine rest as it is classically performed? Is it necessary to compare only shapes recorded in the same condition? A 12-lead ECG, from 14 normal subjects, in three experimental conditions (supine rest, standing and exercise), has been recorded. An analysis based on the computation of the Discrimination Coefficient (DC) between intra-and inter-subjects shows that comparing shapes recorded in the same condition leads to similar values of DC for the three conditions, provided that the interval length is lower than 800 ms. Merging the conditions leads to values of DC always greater than 1 provided that the interval length is lower than 600 ms. A clustering approach, based on correspondence analysis and hierarchical ascending classification, shows that, when merging the conditions, the ECG of a subject are in the same cluster in 94% of the cases.