A recent accelerated SOR-like method for generalised saddle point problems is discussed. Sufficient conditions for convergence are derived, and some numerical experiments illustrate its effectiveness.