Under what circumstances does a theoretical scientific term possess empirical significance? Many proposed criteria of empirical significance can be regarded as precisifications of what I call the ‘novel predictions’-principle, that a term t possesses empirical significance if and only if some hypothesis H(t) involving t, in conjunction with a scientific theory TC in which t occurs, yields an observational prediction not entailed by TC alone. Rozeboom’s (1960) criticism of Carnap’s (1956) criterion of empirical significance amounts in essence to the point that in observationally complete theories – theories which decide every observation sentence – novel predictions are not to be had. This is a serious problem for the ‘novel predictions’-principle. I hold that it provides motivation to radically rethink what a criterion of empirical significance should look like. The criterion I propose is based on the verificationist tenet that a term’s meaningfulness amounts to its being linked to observation terms (in a way to be specified). This is naturally understood as the idea that meaningfulness should be identified with empirical significance. According to the logical empiricists, theoretical terms acquire their meaning via being partially interpreted by a theory. I present a formal criterion for when a theory partially interprets a term and discuss various possible objections to it.