To find out whether the initial orientation of pigeons is affected by a spontaneous directional tendency, as postulated by Wallraff's (1974, 1978, 1982) hypothesis, experienced birds were released at test sites distributed symmetrically around their loft. The length of the mean vectors of the single releases, the deviations from the home direction, the homeward components as well as the homing speed did not show a correlation with the geographic position of the home direction. Summarizing four sites each on 21 experimental circles, we frequently obtained significant compass vectors, but they varied in direction between 115 degrees ESE and 351 degrees N, depending on what sites had been used, and did not indicate a uniform trend. A 'preferred compass direction' as an integrated part of pigeon navigation, being the reason for the frequently observed deviations from the home direction, could not be confirmed. The problematic nature of simply pooling the data of several symmetrically distributed test sites and calling any resulting significant vector a 'preferred compass direction' is discussed, together with other possible reasons for asymmetrical distributions of release site biases.