Robust methods for breath sampling and analysis are required for potential clinical applications. We have evaluated an improved sampling and experimental design, assessing instrumental and biological variability within and across breath measurements. Calibration curves for selected relevant volatile organic compounds (VOCs) were produced to look at the dynamic range and stability of analysis by gas chromatography/time-of-flight mass spectrometry. Linear responses were observed with R 2 > 95% and limits of detection in mid-range pg/μl. Overall experimental design, visualized by means of principal component analysis, demonstrated good clustering on quality control samples, background air and blanks, with dispersion observed as expected across human breath samples. Serial sampling while breathing VOC-filtered air for up to 30 min demonstrated marked variation in reproducibility of VOCs, with median intraclass correlation coefficient of 0.29 (interquartile range 0.16-0.71), with no apparent effect of disease status. We have shown that we can reliably detect VOCs at very low concentrations in exhaled breath samples, and that the reproducibility depends on (a) compound of interest; (b) length of time breathing VOC-filtered air. These parameters will require investigation in studies of potential breath biomarkers, and must be standardized if tests are to become clinically useful. © 2012 IOP Publishing Ltd.