This paper presents a new semantic space creation method with an adaptive axis adjustment mechanism for media data retrieval. The semantic space is essentially required to search semantically related and appropriate information resources from media databases. In the method, data in the media databases are mapped as vectorized metadata on the semantic space. The distribution of the metadata on the semantic space is the main factor affecting the accuracy of the retrieval results. In the method, an adaptive axis adjustment mechanism is used to rotate and combine the semantic correlated axes on the semantic space, and remove axes from the semantic space. We demonstrated by experiments that when the semantic space is created and adjusted based on the semantic correlated factors, the metadata are appropriately and sharply distributed on the semantic space.