This paper proposes a novel Adaptive Compass Search (ACS) for Maximum Power Point Tracking (MPPT) of a centralized Thermoelectric Generation (TEG) system under the condition of Non-uniform Temperature Distribution (NTD). The centralized TEG system employs only one MPPT converter and thus the overall costs of operation and maintenance can be significantly reduced compared to that of string-type and modularized configurations. However, multiple Maximum Power Points (MPPs) usually appear in a centralized TEG system under an NTD condition. In order to effectively and efficiently seek the global MPP, an adaptive sequence of exploration directions is employed to considerably enhance the global searching ability of ACS through the previous searching results. Three case studies are carried out to study the MPPT performance of ACS, including constant temperature difference, step change of temperature, and random temperature variation. Simulation results show that ACS can obtain a high-quality global MPP with a faster speed and a higher convergence stability than those of Perturb and Observe (P&O), Particle Swarm Optimization (PSO), and Compass Search (CS), respectively. This work is supported by National Natural Science Foundation of China (No. 61963020).