An image forming apparatus having a cleaning mechanism for cleaning paper powder or toner pressed and remaining on the surface of the circumference of an endless belt is provided. The cleaning mechanism includes a rotational shaft part that is detachably and rotatably attached to a supporting frame. The supporting frame of the cleaning mechanism is turned with the rotation of the rotational shaft part so that the supporting frame of the cleaning mechanism faces the belt supporting frame, fixing both frames at a point where a cleaning member presses the surface of the circumference of the endless belt.