We demonstrate the effective object search in unknown 3D environment by using Object Co-occurrence Graph (OCG). The proposed OCG is automatically constructed from many word-tagging images provided by millions of Internet users. It describes the co-occurrence relations between objects. Since the small target object is difficult to be recognized due to insufficient image's resolution or occlusion problem, the finding of large objects which relate to particular target object will be more appropriate. This large object is so called cue object since it provides hints about target object's location. In our approach, robot solely depends on visual information; visual grid map and proposed object detectors, for its task. Moreover we also show that the employment of OCG can actually make the searching task become easier, reduce operating time, and minimize the robot's movement trajectory.