It has been widely assumed that both subjects and adjuncts block sub-extraction out of them (Condition on Extraction Domains, CED: Huang 1982). In the recent minimalist framework, unified accounts of the CED have been proposed. For example, Uriagereka (1999) and Nunes and Uriagereka (2000) derive the CED from an LCA-based linearization algorithm (Kayne 1994) that subjects and adjuncts are islands because they need to be 'flattened out' early for linearization, and this consequently renders their sub-constituents inaccessible for further computation. Stepanov (2007), however, empirically argues against the CED, reporting that subjects are not islands in languages such as German and Japanese. If some languages lack subject island effects, any unified analysis of subject and adjunct islands must be abandoned for those languages and thus cannot be a universal account. Using German and Japanese data from three controlled 7-point scale acceptability judgment studies, we present counter-evidence to his claim and argue that the unified accounts of the CED are still tenable.