Head-related transfer functions (HRTFs) or its time-domain counterpart, head-related impulse responses (HRIRs) play a crucial role in 3-D audio rendering. The HRIRs with a couple of hundred-sample lengths result in the high computation cost for the real-time 3-D audio applications especially when multiple sound sources are rendered simultaneously. To overcome this problem, various modeling approaches have been reported to shorten HRIRs. In this research, an efficient finite-impulse-response (FIR) model is studied which is essentially based on the concept of the minimum-phase modeling technique. In this method, the measured HRIRs are represented by the interaural time delay (ITD) and the magnitude spectra that are approximated by two FIR filters. To investigate the accuracy dependence of this modeling approach on the order of FIR filter, two psychoacoustic listening experiments were conducted. In the examinations, two HRIR databases (MIT and CIPIC) are exploited and evaluated in terms of sound source localization and sound quality by comparing the synthesized stimuli with the measured HRIRs and those with the FIR models of different orders. Experimental results indicated that the measured hundred-sample-length HRIRs can be sufficiently modeled by the low-order (a dozen of coefficients) FIR model from the perceptual point of view, which represents an important reduction in the computational cost for implementing real-time 3-D audio applications.