Quasi-optimal observables vs. event selection cuts