This paper proposes a novel approach to automated music recommendation systems. Current systems use a number of methods, although these are generally based on similarity of content, contextual information or user ratings. These approaches therefore do not take into account relevant, well-established models from the field of music psychology. Given recent evidence of this field’s excellent capacity to predict music preference, we propose a function based on both the Ebbinghaus forgetting curve of memory retention and Berlyne’s inverted-U model to inform recommendation systems through “collative variables” such as exposure/familiarity. According to the model, an intermediate level of these variables should generate relatively high preference and therefore presents significant untapped data for music recommendation systems.