Botnets are and are likely to remain the main vehicle for online crime for the foreseeable future. To protect their business models, botnet operators constantly improve their protocols and applications to harden them against detection, analysis and takedown efforts. Our analysis suggest that future botnets will use proper encryption for their protocol messages, rendering them invisible to most deployed network intrusion detection systems. Therefore, we identify properties of botnet communications that will not be obscured by these measures. This allows us to design features which can be derived by measuring these properties and used as input to an approach for detecting systems infected with a botnet client. First measurements on network traffic generated by legitimate applications and a botnet client suggest that our features are capable of reliably discriminating between the two.