High-resolution , high-signal-to-noise ( \langle S/N \rangle = 85 ) spectra have been obtained for five stars – CD–24 ^ { o } 17504 , CD–38 ^ { o } 245 , CS 22172–002 , CS 22885–096 , and CS 22949–037 – having [ Fe/H ] \lesssim –3.5 according to previous lower S/N material .
LTE model-atmosphere techniques are used to determine [ Fe/H ] and relative abundances , or their limits , for some 18 elements , and to constrain more tightly the early enrichment history of the Galaxy than is possible based on previous analyses .

We compare our results with high-quality higher-abundance literature data for other metal-poor stars and with the canonical Galactic chemical enrichment results of Timmes et al .
( 1995 ) and obtain the following basic results : ( 1 ) Large supersolar values of [ C/Fe ] and [ N/Fe ] , not predicted by the canonical models , exist at lowest abundance .
For C at least , the result is difficult to attribute to internal mixing effects ; ( 2 ) We confirm that there is no upward trend in [ \alpha /Fe ] as a function of [ Fe/H ] , in contradistinction to some reports of the behavior of [ O/Fe ] ; ( 3 ) The abundances of aluminum , after correction for non-LTE effects , are in fair accord with theoretical prediction ; ( 4 ) We confirm earlier results concerning the Fe-peak elements that [ Cr/Fe ] and [ Mn/Fe ] decrease at lowest abundance , while [ Co/Fe ] increases – behaviors that had not been predicted .
We find , however , that [ Ni/Fe ] does not vary with [ Fe/H ] , and at [ Fe/H ] \sim –3.7 , [ Ni/Fe ] = +0.08 \pm 0.06 .
This result appears to be inconsistent with the supernova models of Nakamura et al .
( 1999 ) that seek to understand the observed behavior of the Fe-peak elements by varying the position of the model mass cut relative to the Si-burning regions .
( 5 ) The heavy neutron-capture elements Sr and Ba exhibit a large scatter , with the effect being larger for Sr than Ba .
The disparate behavior of these two elements has been attributed to the existence of ( at least ) two different mechanisms for their production ; ( 6 ) For the remarkable object CS 22949–037 , we confirm the result of McWilliam et al .
( 1995b ) that [ C/Fe ] , [ Mg/Fe ] , and [ Si/Fe ] are supersolar by \sim 1.0 dex .
Further , we find [ N/Fe ] = 2.7 \pm 0.4 .
None of these results is understandable within the framework of standard models .
We discuss them in terms of partial ejection of supernova mantles ( Woosley & Weaver 1995 ) and massive ( 200–500 M { { } _ { \odot } } ) zero-heavy-element hypernovae ( e.g .
Woosley & Weaver 1982 ) .
The latter model actually predicted overproduction of N and underproduction of Fe-peak elements ; and ( 7 ) We use robust techniques to determine abundance trends as a function of [ Fe/H ] .
In most cases one sees an apparent upturn in the dispersion of relative abundance [ X/Fe ] for [ Fe/H ] < –2.5 .
It remains unclear whether this is a real effect , or one driven by observational error .
The question needs to be resolved with a much larger and homogeneous data set , both to improve the quality of the data and to understand the role of unusual stars such as CS 22949–037 .