The performance of robust controllers hinges on the underlying model set. The aim of the present paper is to develop a system identification procedure that enables the design of a controller that achieves optimal robust performance. Hereto, the complex interrelation between system identification and robust control is thoroughly analyzed and novel connections are established between (i) control-relevant and coprime factor identification and (ii) model uncertainty size and the control criterion. The key technical results include new robust-control-relevant and (Wu,Wy)-normalized coprime factorizations. The results enable the identification of multivariable model sets that achieve high robust performance in a subsequent robust control synthesis. Superiority of the proposed results compared to existing approaches is shown by means of an example.