A document is mainly composed of words and images, but the complex relationship that binds these two completely different semiotic resources is usually taken for granted as transparent. The simple relations between words and images – ‘anchorage’ and ‘relay’, identified by Barthes almost 30 years ago – are unable to deal with the complexity of their bond, made even more complex by current printing and computer technology. This paper aims to identify the potential relations that bind texts and images together by arguing for a multilevel description of their logico-semantic relationships. The multiple, evaluative and metaphorical functions of the relations will also be discussed. The data generated from the proposed framework can form an empirical corpus for quantitative analysis. Examples from a variety of sources will be used as examples to show how the framework can be operationalized.