Five halo dwarfs in the temperature range of the Population II Li plat
eau and all having [Fe/H] < - 3.0 were found to have mildly depleted L
i abundances, down on average by 0.15 dex, compared to higher metallic
ity halo dwarfs of comparable temperature. Evolutionary models tracing
Li abundances in halo stars in the presence of several Li depletion m
echanisms do not predict this difference. Four comparison stars with -
2.8 < [Fe/H] < - 0.6 were found to be in excellent agreement with pre
vious studies. Our sample also includes the very metal poor dwarf, CS
22876-32 ([Fe/H] = - 4.2), for which we report a lithium abundance in
the range log (n(Li)/n(H)) + 12.00 = 1.8 - 2.0, based on our observati
on of a 15 mA Li line, contrary to a nondetection reported by Molaro,
but in agreement with recent measurements reported by Thorburn & Beers
. We also trace the dependence of Li abundances on stellar effective t
emperatures, amounting to 0.03 dex per 100 K, in agreement with the pr
edictions of depletion models. Observed Li abundances are normalized t
o an effective temperature near the hot end of the Spite plateau, 6200
K, and must be interpreted with this renormalization in mind. Two gen
eral possibilities are considered for the lower Li abundances seen in
stars with [Fe/H] < - 3.0: (1) that a small amount of Li production ha
s occurred during halo formation, in which case the primordial Li abun
dance is near log (n(Li)/n(H)) + 12.00 = 2.05 or lower, or (2) that th
e most metal poor halo stars have depleted their surface Li abundances
in a fashion not predicted by current models, and the primordial Li v
alue lies instead near 2.20. The difference in the implied primordial
value alters the constraints on various cosmological models, most nota
bly the baryon density resulting from the big bang. Until the differen
ce between higher and lower metallicity halo stars is understood, it c
annot be stated with certainty whether observing more metal poor stars
leads one closer toward or further away from the primordial Li abunda
nce.