Difficulty of condition monitoring and diagnosis against machineries constructed by some mechanical elements, which are located close range increases more, because of masking in many frequency bands such as discrete frequency signals associated with shaft speed and broad band ones associated with the natural frequencies of each machinery elements and structural materials. This paper proposes a novel condition monitoring method on rotary machineries by fault vibration signals extraction algorithm based on the proposed multiplex elements Auto Regressive model (MEAR model), which can be removed influence of vibration signal components generated by other machineries located close range except for monitored and diagnosed rotary machinery. We intend to have attempted to demonstrate the effectiveness of the proposal MEAR model and its extraction algorithm using an experimental rotary machine which, is constructed by a rotor, spur gears and rolling element bearings and an induction motor, which are connected to a drive and two driven shafts. The experimental machine can be artificially assumed three common faults such as an unbalanced rotor, a gear unbalance and wear gear teeth. Good results are obtained in this article, the ability of the proposal condition monitoring method (MEAR model and its algorithm) is effectively identified. High precision condition monitoring by the method can be carried out under the environment of close range location.