The localization performance was studied when subjects listened 1) to a real sound field and 2) to artificial-head recordings of the same sound field. The experiments took place in a standard listening room where each stimulus (female speech) was emitted from one of 19 loudspeakers, and the subjects were to indicate the perceived sound source. The artificial-head recordings were made a) by the artificial heads' built-in microphones and b) by blocked ear canal microphones. The reproduction was carried out by carefully equalized headphones. Eight artificial heads were included in the investigation, and 20 subjects participated, except for the experiment with recordings from built-in microphones, which was performed for eight subjects. When compared to real life, the localization performance with the artificial heads resulted in an increased number of errors independent of the recording technique. In general, the directions in the median plane were frequently confused, not only with nearby directions, but also with directions further away. For some artificial heads there was also an increase in confusions of directions outside the median plane. A much better performance is obtainable with binaural recordings made in the ears of humans. This encourages the design and production of improved artificial heads.