The non-individualized head related transfer function (HRTF) is known to have a few problems, which are referred to as the 'hole in the middle' phenomenon and 'front-back reversals.' To overcome these problems, an HRTF refinement technique was introduced, but unfortunately, this refinement technique causes sudden degradation in sound quality and difficulty in cross-talk cancellation because of notch frequency exaggeration. In this paper, an HRTF refinement using directional weighting function is proposed. This newly proposed technique weights ordinary HRTF according to its direction to amplify frontal sound intensity. Since the proposed technique does not exaggerate the notch frequency, spectral differences in the 'cone-of-confusion' region become more pronounced within overall audible frequencies, resulting in mitigating the sound degradation. In addition, the cross-talk cancellation can be done more easily. We verified the superiority of the proposed technique over the existing one by means of the sound localization and sound quality tests in headphone and loudspeakers.