Compensation effect is an unsolved issue when determining pyrolysis kinetics of biomass using inverse modelling and optimization algorithms, implying no unique solution can be obtained. To address this problem, a new method coupling Gauss multi-peak fitting method, Kissinger method, a numerical model, and Shuffled Complex Evolution (SCE) optimization algorithm is proposed to extract kinetics from microscale thermogravimetric analysis (TGA) experiments and avoid attainment of unreasonable good-fit solutions. TGA tests of beech wood at nitrogen atmosphere were conducted at three heating rates. Gauss multi-peak fitting method was employed to separate the overlapped peaks in TGA curves and identify the contribution of each elemental component reaction. The kinetics of individual reactions were estimated by Kissinger method to provide an initial solution for SCE optimization. Then, narrow initial search ranges were determined to further refine the kinetics by SCE. By compiling previous data in literature and our optimization results, it was found compensation effect exists for individual basic components, hemicellulose, cellulose and lignin, following a linear correlation between lnA andEa, and among multiple components. Pyrolysis of hemicellulose can be modelled by either first-order or high-order reactions, while cellulose pyrolysis is more likely a first-order reaction. Nevertheless, the long tail in DTG curves associated with decomposition of lignin can only be captured by a high-order reaction. Although the Kissinger solution of lignin cannot be used in selecting an appropriate search range for SCE optimization, the narrow search ranges of the remaining kinetic parameters can ensure the accuracy and convergency efficiency of optimization.