2010MNRAS.403..797G

High spectral resolution evolutionary synthesis models have become a routinely used ingredient in extragalactic work, and as such deserve thorough testing. Star clusters are ideal laboratories for such tests. This paper applies the spectral fitting methodology outlined in Paper I to a sample of clusters, mainly from the Magellanic Clouds and spanning a wide range in age and metallicity, fitting their integrated light spectra with a suite of modern evolutionary synthesis models for single stellar populations. The combinations of model plus spectral library employed in this investigation are Galaxev/STELIB, Vazdekis/MILES, SED@/GRANADA and Galaxev/MILES+GRANADA, which provide a representative sample of models currently available for spectral fitting work. A series of empirical tests are performed with these models, comparing the quality of the spectral fits and the values of age, metallicity and extinction obtained with each of them. A comparison is also made between the properties derived from these spectral fits and literature data on these nearby, well studied clusters. These comparisons are done with the general goal of providing useful feedback for model makers, as well as guidance to the users of such models. We find the following. (i) All models are able to derive ages that are in good agreement both with each other and with literature data, although ages derived from spectral fits are on average slightly older than those based on the S-colour-magnitude diagram (S-CMD) method as calibrated by Girardi et al. (ii) There is less agreement between the models for the metallicity and extinction. In particular, Galaxev/STELIB models underestimate the metallicity by ∼0.6 dex, and the extinction is overestimated by 0.1 mag. (iii) New generations of models using the GRANADA and MILES libraries are superior to STELIB-based models both in terms of spectral fit quality and regarding the accuracy with which age and metallicity are retrieved. Accuracies of about 0.1 dex in age and 0.3 dex in metallicity can be achieved as long as the models are not extrapolated beyond their expected range of validity.