This paper describes the challenges that are being tackled and those that remain to be addressed if we are to enable electronic health record information to be shared seamlessly and meaningfully. This goal is known as semantic interoperability, and is needed if computational services are to be able to interpret safely clinical data that has been integrated from diverse sources. Based on sustainable architectural approaches, the paper describes the clinical case for consistently expressed clinical meaning within electronic health records, in particular where computers rather than humans need to be able to process EHR data safely. It outlines the main kinds of information and knowledge artefact that are used to represent meaning within EHRs, and considers for each its role and limitations. The problems that arise with trying to use terminology consistently with EHR reference models is explored, together with the implications for designing EHR archetypes. Examples are given of situations where a diversity of options exists for how to represent compound (multi-part) clinical expressions. Recommendations are made for the kinds of change that are needed both in record structures and in terminology systems to minimise this diversity and thereby aid semantic interoperability.