This paper presents an original approach to simply and efficiently estimate the center of mass position of a free-floating base system, like a humanoid robot or a human body. This approach is based on the theory of complementary filtering, which is a popular technique in aerial robotics, but rarely implemented in humanoid robotics. The main idea consists in merging input measurements like the zero-moment point position, the contact forces, etc., which are then filtered according to their reliability in their respective spectral bandwidth. We validate this approach in simulation by (i) comparing the estimated center of mass trajectory with its real value and (ii) showing that the complementary filter offers on average a least reconstruction error than the classic Kalman filter.