The pyrolysis and combustion behaviour of a set of 11 Canadian coals with different ranks (lignite to low volatile bituminous) and maceral compositions has been investigated by TGA techniques. Temperature programmed heating of the coals was carried out both under nitrogen and under air, and the weight loss rates occurring in the two experimental conditions were compared in the whole temperature range studied (100±10008C). Results showed that the pyrolysis curves of the coals do not match at all with any speci®c feature of the corresponding combustion pro®les, and that the temperatures of initiation of both processes are very different in the low-rank end (higher initiation temperatures for pyrolysis), to become similar only for coal ranks of ,0.8% vitrinite re¯ectance and above. This contradicts existing theories which state that coal combustion under TGA conditions is a three-stage process, namely volatiles release, vitrinite combustion and inertinite combustion. The processes leading to the weight loss rates occurring in the early stages of combustion were also investigated, with special emphasis in the temperature interval where no substantial weight losses had occurred yet in either a combustion or a pyrolysis experiment. This was done by heating the coals to 3008C in the TGA under air, and then switching the gas ¯ow to nitrogen and allowing the sample to further devolatilise until 10008C. Also, partly burnt and/or pyrolysed samples were obtained from the TGA and characterised by optical microscopy techniques. It was observed that the volatile yields of all the coals were substantially reduced as a consequence of their initial heating under air. Besides, evidences of melting and thermal annealing in the inner core of burning coal particles were noticed to occur at lower temperatures than in pyrolysing particles. This was attributed to a sealing effect of the oxidation rim formed in the early stages of combustion, which might give rise to a higher pressure build-up in the inner part of burning particles, thus enhancing the likelihood for condensation reactions to take place in the newly formed metaplast. As combustion pro®les are commonly used to infer about combustion behaviour of coals, much care should be exercised in interpreting them, since even in a pure vitrain, two rather than one single material will be involved in the measured weight losses, and, more, these materials will often display fairly different reactivities.