This paper examines year‐on‐year changes to the composition of performance peer groups used for relative performance evaluation in setting CEO pay in FTSE 100 companies and finds evidence of peer selection bias. The authors find that firms keep their peer groups weak by excluding relatively stronger performing peers. They also show that peer selection bias is less pronounced in firms with higher institutional investor ownership, which suggests that institutional investors might be aware of the risks of peer selection bias. The results suggest that peer group modifications can be viewed, at least in part, as an expression of managerial rent‐seeking.