Methods of recording landmarks on the facial skin without the use of markers have become increasingly accepted in image-guided surgery. However, position or muscular activity may change the skin's geometry and generate a lack of agreement between the facial contours recorded before and those recorded during the operation. In the present study, we measured this physiological shift of facial skin and evaluated its influence on the accuracy of stereotactic recording. We made laser-scans of the skin of 20 conscious patients while they were sitting and lying, both at rest and when smiling. The laser-scans were referenced to the corresponding computed tomographic dataset, and the accuracy of the recording was calculated. Gravitational or muscular shifts of the skin reduced the mean (S.D.) accuracy of recording to 1.7 (0.3)mm. The loss of accuracy was significantly correlated with the dynamic and gravitational wrinkling of the facial skin and with the body mass index of each patient.