Dear Experts, I'm a new R user and I'll appreciate your help regarding the following. I'm trying to generate an exhaustive search of all candidate models in a simple linear regression and select the one with the lowest CV-error (or alternatively the lowest Error on a Test set -- if I have lots of data). The leaps package can generate this exhaustive search but all models are evaluated on the train data (without cross-validation). How can I implement what I'm trying to achieve? Any guidance will help... library(ElemStatLearn) #Follow the example of Page 58 in Elements of Stat Learning Book train <- subset(prostate, train==TRUE )[,1:9] test <- subset(prostate, train=FALSE )[,1:9] #Best subset selection library(leaps) prostate.leaps <- regsubsets( lpsa ~ . , data=train, nbest=100, nvmax=8, method="exhaustive", really.big=T) prostate.leaps.sum <- summary(prostate.leaps) prostate.models <- prostate.leaps.sum$which prostate.models prostate.models.rss <- prostate.leaps.sum$rss prostate.models.rss prostate.models.size <- as.numeric(attr(prostate.models, "dimnames")[[1]]) prostate.models.best.rss <-tapply(prostate.models.rss, prostate.models.size, min) prostate.models.best.rss Thanks a lot! Lars. [[alternative HTML version deleted]]