We have used moderately high-resolution , high S/N spectra to study the chemical composition of 10 field ab -type RR Lyrae stars .
Variables having accurate photometric and radial velocity data were selected , in order to derive a precise estimate of the atmospheric parameters independently of excitation and ionization equilibria .
A new temperature scale was determined from literature Infrared Flux Method measures of subdwarfs and the Kurucz ( 1992 ) model atmospheres , and used to calibrate colors for both dwarfs and RR Lyraes .
Photometric reddening estimates for the program stars were carefully examined , and compared with other determinations .
The applicability of Kurucz ( 1992 ) model atmospheres in the analysis of RR Lyraes at minimum light was analyzed : we found that they are able to reproduce colors , excitation and ionization equilibria as well as the wings of H _ { \alpha } .
The comparison solar abundances were carefully determined .
From a new analysis of weak Fe I lines with accurate gf s ( Bard & Kock 1994 ) we derived \log \epsilon ( Fe ) _ { \odot } = 7.52 , in agreement with the Fe abundances determined from meteorites and Fe II lines .

We derived abundances for 21 species .
Main results are :

• The metal abundances of the program stars span the range -2.50 < [ Fe/H ] < +0.17 .

• Lines of most elements are found to form in LTE conditions .
Fe lines satisfy very well the excitation and ionization equilibria .
A comparison with statistical equilibrium computations shows that rather large collisional cross sections are required to reproduce observations .
If these cross sections are then used in the analysis of the formation of Fe lines in subdwarfs and RGB stars , no significant departures from LTE are found for these stars , thus validating the very numerous LTE analyses .

• RR Lyraes share the typical abundance pattern of other stars of similar [ Fe/H ] : \alpha -elements are overabundant by \sim 0.4 dex and Mn is underabundant by \sim 0.6 dex in stars with [ Fe/H ] < -1 .
Solar scaled abundances are found for most of the other species , except for the low Ba abundance in the extremely metal-poor star X Ari ( [ Fe/H ] \sim - 2.5 ) .

• Significant departures from LTE are found for a few species : Nd II , Ce II , Y II and Sc II are severely underabundant ( \sim 0.5 dex ) in metal-rich variables ; Ti I and Cr I are slightly ( \sim 0.1 - 0.2 dex ) underabundant in metal-poor stars .
These effects are attributed to overionization .
We suggest that the photoionization of the alkaline earth-like ions is due to Lyman lines emission produced by the shock waves that propagate in the atmosphere of these variables ( Fokin 1992 ) .

• Departures from LTE were considered in detail in the derivation of abundances for the light elements ( O and Na ) .
Significant corrections were required for the O I IR triplet and the Na D lines .
The resulting pattern reproduces that observed in less evolved field stars .
We did not find any evidence for an O-Na anti-correlation among these field HB-stars , suggesting that the environment is likely to be responsible for the anti-correlation found in metal-poor globular cluster stars ( Sneden et al 1992 ) .

We used our new [ Fe/H ] abundances , as well as values from Butler and coworkers ( corrected to our system ) , and from high resolution spectroscopy of globular clusters giants , to obtain a revised calibration of the low-resolution metallicity index \Delta S ( Preston 1959 ) :

{ [ Fe / H ] } = -0.194 ( \pm 0.011 ) \Delta S - 0.08 ( \pm 0.18 ) Our new metallicity scale is stretched on both low and high metallicity ends with respect to Butler ’ s ( 1975 ) .
The error in [ Fe/H ] by \Delta S observations is 0.16 dex , well of the same order of high resolution metallicity determinations .
The slope of the calibration obtained considering only stars with 4 < \Delta S < 10 is slightly smaller than that obtained using all stars .
While this difference is only barely significant , it might point out the presence of a non-linearity of the \Delta S vs [ Fe/H ] relation , as suggested by Manduca ( 1981 ) .

The new [ Fe/H ] values were used to update the metallicity calibration of the Ca II K line index ( Clementini et al 1991 ) .
Using the present new metallicities , and W ^ { \prime } ( K ) values and relative errors from Clementini et al ( 1991 ) , a least-squares fit weighted both in W ^ { \prime } ( K ) and [ Fe/H ] gives :

{ [ Fe / H ] } = 0.65 ( \pm 0.17 ) W ^ { \prime } ( K ) -3.49 ( \pm 0.39 ) Finally , our new metallicity scale was used to revise the metallicity dependence of the absolute magnitude of RR Lyrae stars , M _ { V } .
Using M _ { V } values from Fernley ( 1994 ) for the field stars , and estimates from Liu & Janes ( 1990b ) and Storm et al ( 1994 ) for the cluster variables , we found :

{ M } _ { V } = 0.20 ( \pm 0.03 ) { [ Fe / H ] } +1.06 ( \pm 0.04 ) and :

{ M } _ { V } = 0.19 ( \pm 0.03 ) { [ Fe / H ] } +0.96 ( \pm 0.04 ) the last being obtained by using M _ { V } estimates derived for a value of the conversion factor between observed and true pulsation velocity p = 1.38 ( Fernley 1994 ) .
The adoption of the new metallicity scale does not yield significant changes in the slope and zero-point of the M _ { V } vs [ Fe/H ] relation .
Observations do not rule out the possibility that the slope of the M _ { V } vs [ Fe/H ] relation might be different for metal-poor and metal-rich variables .
However , a larger sample of Baade-Wesselink M _ { V } determinations is requested to definitely settle this question .