A good knowledge of both accuracy and precision of the Hipparcos paral
laxes is one of the keys for their future scientific use. For this pur
pose, the Hipparcos preliminary parallaxes, as obtained after the proc
essing of the first 30 months of Hipparcos data, are compared to vario
us ground-based parallax estimates, using astrometric, photometric and
spectroscopic data. In order to find unbiased values of the global ze
ro-point and of external errors, a new maximum-likelihood algorithm ha
s been built, taking into account the censorships of the observed data
. Applying the method to a sample of distant stars? it is shown that t
he global zero-point error of the Hipparcos preliminary parallaxes sho
uld be smaller than 0.1 mas and that the external errors cue unlikely
to be underestimated by more than about 5%.