We have performed accurate iron abundance measurements for 44 red giants ( RGs ) in the Carina dwarf spheroidal ( dSph ) galaxy .
We used archival , high-resolution spectra ( R \sim 38 000 ) collected with UVES at ESO/VLT either in slit mode ( 5 ) or in fiber mode ( 39 , FLAMES/GIRAFFE-UVES ) .
The sample is more than a factor of four larger than any previous spectroscopic investigation of stars in dSphs based on high-resolution ( R \geq 38 000 ) spectra .
We did not impose the ionization equilibrium between neutral and singly-ionized iron lines .
The effective temperatures and the surface gravities were estimated by fitting stellar isochrones in the V , B–V color-magnitude diagram .
To measure the iron abundance of individual lines we applied the LTE spectrum synthesis fitting method using MARCS model atmospheres of appropriate metallicity .
For the 27 stars for which we measured both Fe I and Fe II abundances , we found evidence of NLTE effects between neutral and singly-ionized iron abundances .
The difference is on average \sim 0.1 dex , but steadily increases when moving from the metal-rich to the metal-poor regime .
Moreover , the two metallicity distributions differ at the 97 % confidence level .
Assuming that the Fe II abundances are minimally affected by NLTE effects , we corrected the Fe I stellar abundances using a linear fit between Fe I and Fe II stellar abundance determinations .
We found that the Carina metallicity distribution based on the corrected Fe I abundances ( 44 RGs ) has a weighted mean metallicity of \hbox { [ Fe / H ] } = -1.80 and a weighted standard deviation of \sigma = 0.24 dex .
The Carina metallicity distribution based on the Fe II abundances ( 27 RGs ) gives similar estimates ( \hbox { [ Fe / H ] } = -1.72 , \sigma = 0.24 dex ) .
The current weighted mean metallicities are slightly more metal poor when compared with similar estimates available in the literature .
Furthermore , if we restrict our analysis to stars with the most accurate iron abundances , \sim 20 Fe I and at least three Fe II measurements ( 15 stars ) , we found that the range in iron abundances covered by Carina RGs ( \sim 1 dex ) agrees quite well with similar estimates based on high-resolution spectra .
However , it is a factor of two/three smaller than abundance estimates based on the near-infrared Calcium triplet .
This finding supports previous estimates based on photometric metallicity indicators .