An audio radiation type reflective sound box structure comprises a hollow cone tube having a first end received in a box, a second end extended out of an opening of the box to form an inclined side wall, and at least one vent disposed adjacent to the first end. When a diaphragm of a speaker mounted on an inner side edge of the second end generates vibrations, compressed air in the hollow cone tube is pushed out of the opening through the vent, and then collides with the outer edge of the inclined side wall and is diffused toward a peripheral direction adjacent to a front side of the speaker along an extension direction of the inclined side wall. Thus, the sounds reflected by the box can be evenly transmitted to a peripheral environment adjacent to the front side of the speaker.