Lawrence and Rugh (1995, Automatica 31, 381-390 and, Kaminer et al. Automatica 31, 1185-1191), propose that the realisation for a gain-scheduled controller should be chosen to satisfy a local linear equivalence condition. However, this provides an inadequate basis for selecting an appropriate realisation. Many realisations satisfy the local linear equivalence condition yet are not equivalent and can exhibit quite different dynamic behaviour. Furthermore, the condition imposes restrictions on the controller states and inputs which are not a priori necessary.