We adopt the usual spectroscopic notation, i.e. 12#12 for any abundance quantity X, and log 13#13 (14#14 for absolute number density abundances.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

IRAF is distributed by the National Optical Astronomy Observatories, which is operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...is

We will indicate with the subscript CG96 the sample of our observed program stars and simply CG the results obtained using the whole sample (observations plus literature data) analyzed in the present paper.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...lines

Throughout this paper, the symbol 33#33 will indicate the standard deviation of a single measurement, while the value after the symbol 34#34 will refer to the standard deviation of the mean.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...K92.

The model atmospheres considered in this paper are those in Kurucz's CD-ROM 13. In these models, convection description include an approximate consideration of possible overshooting in the stellar atmospheres. This point is discussed at length in Castelli et al. (1996), who conclude that at present is not possible to establish whether or not this treatment of convection should be preferred to a more traditional approach without any overshooting. However, this uncertainty is only of minor concern for the metallicity scale established in this paper, insofar the approach used for the Sun and the globular cluster giants is the same.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...present

Expected line strengths were used when determining the microturbulent velocity, following the prescription of Magain (1984).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...columns)

All stars that were in more than one data sets have been independently re-analyzed, and then their [Fe/H] values averaged before computing the cluster's mean [Fe/H].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...here)

Following the non-LTE analysis of Clementini et al. 1995, no significant departures from LTE are to be expected in RGB stars. The differences found in Fe abundances from Fe I and Fe II lines are thus likely to be interpreted as due to errors in the analysis and in the atmospheric parameters.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...tied

As a generic estimate, an error of 100 K in 126#126 translates into a 0.3 dex error in 127#127, when deriving spectroscopically the gravities.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...effect.

Our conclusions would not change had we used M 68 ([Fe/H]=-1.99 on our scale) instead of M 15 as OoII template. While M 68 has a small and well determined reddening (E(B-V)=0.03), M 15 has a better populated instability strip.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...variables.

The reader should be aware of the small inconsistency due to our use of the CFP scale for globular cluster (cooler) giants and the new temperature scale for RR Lyraes; however, as discussed in Gratton et al. (1996), differences between the two scales are well within the intrinsic uncertainties of both scales and do not alter significatively our conclusions.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.