Abstract This work designs and validates a spindle error motion separation technique having a sub-nanometre measurement uncertainty. This technique overcomes typical measurement error sources arising from sensor, indexing or the repositioning of the artifact. We compare and assess various known reversal and multiprobe techniques by means of a novel error analysis method. From this, we develop an improved implementation of the multiprobe technique, which by-passes accurate indexing of the artifact and sensor(s) during testing, as well as unequal sensor sensitivities, in case multiple sensors are used. This is achieved by measuring the error motion consecutively under three different orientations by rotating the stator of the spindle utilising a high-precision indexing table. These modifications result in a measurement uncertainty that is four times smaller than the conventional multiprobe technique. Furthermore, the suppression of the low-order harmonics is reduced by an optimisation of measurement angles. Finally, several experimental tests are performed to quantify the measurement uncertainty and the influence of the measurement angles on the error separation. Repeatability tests on the radial error motion of an aerostatic rotary table show a measurement uncertainty of 0.455nm.