Model atmospheres appropriate for each star were extracted from the grid of
K92.
Values of effective temperature 
and surface gravity 
for giants
observed in 47 Tuc, NGC 6752 and NGC 6397 (sample CG96) were taken from Frogel
et al. (1981, 1983). Frogel et al's 
's are based on visual-near infrared
colours (mainly V-K), transformed into temperatures based on the Cohen et al.
(1978; hereinafter CFP) scale. 
values are determined from the position
of the stars in the colour-magnitude diagram (CMD). These values are considered
as accurate as 
100 K in 
and 
0.3 dex in log g, including
possible uncertainties in the effective temperature scale, as well in the
adopted red giant mass, in the bolometric corrections and in the cluster
distance modulus.
and 
values for stars in the SKPL sample are also generally based on
CFP's scale, except for giants in M 92 and M 15; in these two cases the authors
used 
's from Carbon et al. (1982) and
Trefzger et al. (1983).
The reader should be warned that final values for 
and log g, in the
SKPL sample, are slightly different from the actual values on the
CFP's scale. This is because SKPL adjusted the adopted photometric
temperatures until a slope close to zero was achieved for the relationship
between Fe I abundances from individual lines and excitation potential.
However, as discussed in their papers, the agreement of photometric and final
spectroscopic 
was from the beginning
well within the uncertainties of the temperature
scale itself, so that neither the abundance ratios for individual stars nor the
mean values for the clusters are significatively changed by adopting the
tabulated values of atmospheric parameters from SKPL.
In the G8689 sample, 
and 
values were homogeneously adopted
from Frogel et al. (1979, 1981, 1983) and are also used in the present work.
For other data sets (McW92, NDC) we were able to use atmospheric parameters
from Frogel and coworkers. Only half of the 18 stars from M93 have 
\
and 
values listed in Frogel et al.; for the missing stars, we used
their original values, derived spectroscopically from the dependence of
\
on excitation potential and from the Fe ionization equilibrium. In fact, the
parameters adopted by M93 are with very good approximation on the CFP's scale:
the mean for the stars in common are:
(
=30 K, 8 stars) and log g(Frogel)
(M93)=
dex (
=0.17 dex, 7
stars). For further details, see the original papers
of G8689 and SKPL. The starting input values for [Fe/H]
were those from the original analyses.
Metallicities were obtained varying the metal abundance [A/H] of the model
until it was equal to the derived [Fe/H] value. For the microturbulent
velocity 
, the input values were changed until no trend in Fe abundance
with the EW of Fe I lines was present. However, for giants in M 92 and M 15 only a few lines were
available, so we took the values of 
from the relationship:
derived elsewhere (Carretta et al. 1996;
Gratton & Carretta 1996)
for field stars. The adopted atmospheric parameters are listed in
Table 9 (also available in electronic form) for all the stars
studied in the present paper. Metallicities were
obtained varying the metal abundance [A/H] of the model until it was equal to
the derived [Fe/H] value.