In the present work, thermogravimetric analysis was employed to investigate the interactions between oil shale and its semi-coke during co-combustion process. During the test, the blends of semi-coke and oil shale were prepared at different blending ratios of 1:0, 4:1, 3:1, 2:1, 1:1, and 0:1. The results indicated some interactions were detected between oil shale and semi-coke during the tests. The rapid combustion of organic matter in oil shale, which led to the fuel particle temperature's rapid increase and made semi-coke ignite in advance, improved the co-combustion characteristics in terms of ignition temperature (Ti), the temperature reaching the maximum mass loss rate (T(max)), the maximum mass loss rate (R(max)), the combustibility index (C) and the specific reactivity of the co-combustion. With the increase in semi-coke mass fraction in the blends, the specific reactivity of the samples was found to decrease gradually due to lower volatile content and high carbon condensation structure in semi-coke. As the temperature increased, the specific reactivity of the samples first gradually increased, then decreased attributed to the oxidation of selectivity. The poor diffusion caused mainly by the ash shells made the decomposition of minerals more complicated. It was also analyzed that the 'particle groups' phenomenon that the dispersed particles are agglomerated by some forces in semi-coke ash went against the decomposition of minerals. Activation energies E from distributed activation energy (DAEM) slowly decreased at the initial stage, then increased sharply over a narrow conversion region, which indicted a difficult burnout stage. Meantime, a power law model was employed to investigate quantitatively the interactions. The experimental reaction orders could be predicted accurately from the calculated values. At the third stage, the kinetic parameters E(3) and A(3) were almost as much as the calculated values. At the low temperature region (470-540 degrees C) there was little difference between kinetic parameters E and A from the experiment data and the calculated data. (c) 2009 Elsevier Ltd. All rights reserved.