Localization of real and virtual sound sources was studied using two tasks. In the first task, subjects had to turn their head while the sound was continuously on and press a button when they thought they faced the source. In the second task, the source only produced a short sound and the subjects had to indicate, by pressing one of eight buttons, in which quadrant the source was located, and whether it was located above or below the horizontal plane. Virtual sound sources were created using head-related transfer functions (HRTFs), measured with probe microphones placed in the ear canals of the subjects. Sound stimuli were harmonic signals with a fundamental frequency of 250 Hz and an upper frequency ranging from 4 to 15 kHz. Results, obtained from eight subjects, show that localization performance for real and virtual sources was similar in both tasks, provided that the stimuli did not contain frequencies above 7 kHz. When frequencies up to 15 kHz were included, performance for virtual sources was, in general, poorer than for real sources. Differences between results for real and virtual sources were relatively small in the first task, provided that individualized HRTFs were used to create the virtual sources, but quite large (a factor of 2) in the second task. The differences were probably caused by a distortion of high-frequency spectral cues in the HRTFs, introduced by the probe microphone measurement in the ear canal. (C) 1995 Acoustical Society of America.