Two acceptability judgment experiments were conducted to investigate the role of similarity in weak islands environments. We tested predictions stemming from a narrow ap-proach to similarity (Featural Relativized Minimality) and from a broad approach to similarity (Cue- based memory model) in generating intervention effects. According to Featural Relativized Minimality, only syntactic features triggering movement modulate similarity amongst elements, while according to the Cue- based memory model both syntactic and semantic features modulate similarity. Results show significant effects of both animacy and reversibility of thematic roles, two features that do not trigger movement. Therefore, our results support the broad approach to similarity.