The rapid, accurate and non-invasive diagnosis of respiratory disease represents a challenge to clinicians, and the development of new treatments can be confounded by insufficient knowledge of lung disease phenotypes. Exhaled breath contains a complex mixture of volatile organic compounds (VOCs), some of which could potentially represent biomarkers for lung diseases. We have developed an adaptive sampling methodology for collecting concentrated samples of exhaled air from participants with impaired respiratory function, against which we employed two-stage thermal desorption gas chromatography-differential mobility spectrometry (GC-DMS) analysis, and showed that it was possible to discriminate between participants with and without chronic obstructive pulmonary disease (COPD). A 2.5 dm(3) volume of end tidal breath was collected onto adsorbent traps (Tenax TA/Carbotrap), from participants with severe COPD and healthy volunteers. Samples were thermally desorbed and analysed by GC-DMS, and the chromatograms analysed by univariate and multivariate analyses. Kruskal-Wallis ANOVA indicated several discriminatory (p < 0.01) signals, with good classification performance (receiver operator characteristic area up to 0.82). Partial least squares discriminant analysis using the full DMS chromatograms also gave excellent discrimination between groups (alpha = 19% and beta = 12.4%).