The problem of determining the noise level in any measuring system remains urgent. Radiostereometric analysis (RSA) is a radiographic system of unique accuracy which has applications in areas where minute motions or no motion at all occurs. Examples are micromotion between endoprostheses and bone and in fracture healing. We have determined the accuracy of the RSA system as applied to a clinical series of spinal fusions, where the conditions for RSA were not optimal. Using the usual test-retest methodology on a phantom, we showed that its accuracy can be grossly overestimated in the individual case. We found considerable variations in the accuracy in the individual case, depending on the rigid-body configuration. The overall accuracy, expressed as 3-D "vectors" for rotation and translation, respectively, correlated with the condition number, a method for characterizing the marker configuration. Indeed, the condition number explained as much as 92% of the variation in overall rotation. This condition number, however, cannot be used to analyze the accuracy of one degree of freedom of rotation alone. Mathematical simulation of the accuracy in the individual case of the individual dimension, using in-house software, showed that the accuracy (95% confidence) varied between 0.4 and 4.6 degrees of rotation about the transverse axis, corresponding to a clinical stress series of extension and flexion.