Aims: Signal-averaged electrocardiogram (SAECG) is a non-invasive technique based on the average of multiple QRS complexes to reduce noise without the distorting effects of filtering on the averaged QRS (SAQRS). This study aims at reporting the performances of SAECG in terms of noise reduction and SAQRS modification testing various sources of signal perturbation. Method: A comparative analysis of four sources of perturbation, 1) uncorrelated noise, 2) beat alignment, 3) physiological variability and 4) respiratory movement, was performed. The first two cases were assessed using a computer model of a ventricular beat. The other two cases were assessed using respectively high resolution body surface signals recorded from a torso tank (N=2) and from patient data (N=5). The performances of SAECG were measured using three indicators: i) the root mean square error of the noise (RMSE_noise), ii) the RMSE of the SAQRS (RMSE_SAQRS) to measure the amplitude shift of the SAQRS and iii) the correlation coefficient of the SAQRS (CorrCoeff_SAQRS) to evaluate the distortion of the SAQRS. Results: In the first case, SAECG successfully removed high level of noise, made up of the combination of a white Gaussian noise (WGN) with σ=30 mV and a 50 Hz noise with a signal to noise ratio (SNR) of 9dB since the RMSE_noise equals 1±0.01 mV. The RMSE_SAQRS is slightly damaged by physiological variability (RMSE_SAQRS=7.2±2.5 mV) while comparing five SAQRS resulting from the average of 100 beats each, along the same recording. SAQRS are highly distorted by respiration artefacts with a RMSE_SAQRS≥14 mV and a CorrCoeff_SAQRS ≤ 0.85. Conclusion: SAECG can efficiently de-noise signals in presence of uncorrelated noise without distorting the SAQRS. However, physiological variability and respiration motion introduce distortion and amplitude shift between SAQRS.