Breath gas analysis is a promising technology in the frame of medical diagnostics. By identifying disease-specific biomarkers in the breath of patients, a non-invasive and easy method for early diagnosis or therapy monitoring might be developed. However, to verify this potential and develop diagnostic tools based on breath gas analysis one essential prerequisite is a low variability in measurement of exhaled volatile organic compounds. Therefore, a study has been undertaken in order to identify possible artefacts within the application of a breath gas test in practice, for which the breath gas is analysed by proton transfer reaction-mass spectrometry (PTR-MS). After validating the low instrumental variability by repeatedly measuring standard gas, the variability of breath gas sampling has been evaluated. The latter has been carried out by measuring single breath gas samples (mixed expiratory breath) collected over different periods of time such as 1 min (10 volunteers, 4 breath gas samples each), 1 h (10 volunteers, 11 breath gas samples each) and several days (11 volunteers, 10 breath gas samples each). The breath gas samples were collected in Teflon bags and consecutively measured with PTR-MS. It was found that those samples collected within 1 min and 1 h show a low variability. This was, however, not the case for samples being collected over longer periods of time (15–70 days). Under these circumstances, many volatile organic compounds (VOCs) showed significant day-to-day variation in concentration, although the breath collection had been performed under the same conditions (similar sampling time, sampling technique, sample storage time, measurement conditions, etc). This large variation might be assigned to the influence of room air VOCs, which have been investigated in this work, or with other parameters which will be discussed. It was also found that the variability in the measurement of exhaled concentrations of methanol, acetone and isoprene within different individuals (inter individual variability) is much higher than differences in the same volunteer (intra individual variability) measured over a longer time interval.