Just wondered if there is any chance of increasing the limit from 20? I realise just chucking loads of variables at a machine learning technique is not the right way to go about things, but I find using Signal0 in conjunction with the SIGNALS method a fantastically efficient way of assembling a set of features. I can then open the resulting CSV file in R and quickly test those features with RFE, stabsel, Boruta etc etc. No worries if not possible - even with the 20 limit it's much more efficient than cobbling it all together in R! Thx