This paper discusses the residuals in right ascension and declination
from the Bordeaux fully automated transit circle, corresponding to a t
wo years period. It is shown that the solution can be improved by addi
ng statistically significant terms to the reduction model. The correla
tion between the residuals, considered as a sample of a stochastic sta
tionary process, displays a weak dependence on the declination differe
nce and may be treated as a function of a single variable, the right a
scension lag. The characteristics of the autocorrelation curves can be
interpreted as derived from a power spectrum with two components: a l
ow frequency power law and a flat spectrum. A fitting gives S(nu) = 20
10(-6)nu-1.37 + 1.3 and S(nu) = 89 10(-6)nu-1.37 + 2.0 for the right
ascension and declination residuals, respectively, where the frequency
unit is Hz and the power spectrum is expressed in ('')2/Hz. The expre
ssions are valid in the frequency interval from 3 10(-5) to 3 10(-2) H
z. These spectra are compatible with Hog (1968). It is shown that a pa
rt of the low frequency spectrum can be removed by the adoption of a r
eduction model which includes empirical higher order terms.