The bias drift of a micro-electro-mechanical systems (MEMS) accelerometer suffers from the 1/f noise and the tem-perature effect. For massive applications, the bias drift urgently needs to be improved. Conventional methods often cannot ad-dress the 1/f noise and temperature effect in one architecture. In this paper, a combined approach on closed-loop architecture modification is proposed to minimize the bias drift. The modulated feedback approach is used to isolate the 1/f noise that exists in the conventional direct feedback approach. Then a common mode signal is created and added into the closed loop on the basis of modulated feedback architecture, to compensate for the temperature drift. With the combined approach, the bias instability is improved to less than 13 μg, and the drift of the Allan variance result is reduced to 17 μg at 100 s of the integration time. The temperature coefficient is reduced from 4.68 to 0.1 mg/°C. The combined approach could be useful for many other closed-loop accelerometers.