In engineering and industrial applications, disturbances acting on a dynamic system presents a serious problem to system performance. One way to deal with unknown disturbance is to construct a disturbance observer (DOB) to on-line estimate the disturbance. However, conventional DOB utilizes the system's inverse model, hence, can apply to minimum-phase systems only. This paper proposes an Adaptive DOB, which is based on the adaptive inverse control structure and uses the least-squares algorithm (with co-variance reset) to adaptively estimate the unknown disturbance. An important advantage of the proposed Adaptive DOB is that it applies to both minimum-phase and non-minimum phase systems.