Approximate Bayesian Computation is a relatively recent technique that handles models with intractable likelihoods. It has become the method of choice in population genetics, with a growing corpus of theoretical validations and an equally growing library of dedicated softwares. When the dimension of the data is large, summary statistics need replace the entire data and the choice of these statistics has bearings on the validation of the resulting inference. Those statistics are often used in ABC algorithms without consistency checks. We have derived conditions on summary statistics for the corresponding Bayes factor to asymptotically select the true model. Those conditions, which amount to the expectations of the summary statistics to asymptotically differ under both models, are quite natural. We will also discuss extensions to automated summary statistic selection toward model choice and estimation using random forests.