In a recent paper entitled “Truth does not explain predictive success” , Carsten Held argues that the so-called “No-Miracles Argument” for scientific realism is easily refuted when the consequences of the underdetermination of theories by the evidence are taken into account. We contend that the No-Miracles Argument, when it is deployed within the context of sophisticated versions of realism, based on the notion of truthlikeness , survives Held’s criticism unscathed.