This study describes a non-separative headspace solid phase microextraction-mass spectrometry (HS-SPME-MS) approach, in view of its application to online monitoring of a roasting process. The system can quickly provide representative and diagnostic fingerprints of the volatile fraction of samples and, in combination with appropriate chemometric pattern recognition and regression techniques, can successfully be applied to characterize, discriminate, and/or correlate patterns with the roasting process. Eighty coffee samples of different varieties, geographical origins, and blends were analyzed. The experimental HS-SPME-MS results show that the TIC fingerprint can be used to discriminate the degree of roasting; diagnostic ion abundance(s) or ratios were closely correlated with the roasting process; both could successfully be used as markers or analytical decision makers, to monitor roasting processes online, and to define quality and safety of roasted coffee.