We analyze a general multiposition comparator measurement procedure th
at leads to partial removal of artifact error for a class of problems
including roundness metrology, measurement of radial error motions of
precision spindles, and figure error metrology of high-accuracy optica
l components. Using spindle radial error motion as an explicit example
, we present a detailed analysis of a complete test with N orientation
s of a test ball with respect to the spindle. In particular, we show t
hat (1) all components of the ball roundness error average to zero exc
ept those with frequencies of kN cycles/revolution, where k is a posit
ive integer; and (2) the combined standard uncertainty of the measurem
ent is proportional to 1/root N. We then show how a complete set of me
asurements for an N-position test can be synthesized from only two mea
surements, and we derive a general expression for the combined standar
d uncertainty as a function of the number of positions n (2 less than
or equal to n less than or equal to N) actually measured in an N-posit
ion test. This uncertainty can serve as a useful guide to measurement
design, involving trade-offs between multiple setup cost and complexit
y and required levels of angular harmonic resolution and combined stan
dard measurement uncertainty. Published by Elsevier Science Inc.