Key to the successful operation of any distributed computing system is that messages between systems be interpreted in such a way that the sender's desired effect is achieved. To achieve this effect the meaning of a message as interpreted by the receiver must align with the meaning intended by the sender. If it does not then we have semantic mismatch or variance, which if severe can result in system failure. A system with little semantic variance is a semantically reliable system. While relevant to all heterogeneous distributed systems, semantic reliability is particularly relevant to those systems that seek to utilise the ideas embodied in the notion of the semantic web. These systems include: web services, heterogeneous multi-agent systems, and autonomic computing. In such non trivial systems achieving semantic reliability is a complex and multi-facetted challenge. This paper first considers classic notions of reliability in hardware and software engineering. The problem of meaning exchange is then considered at length from a semiotic standpoint. It is argued that while the use of ontologies certainly effectively mitigates the problem, ontologies alone cannot be the complete solution and human dimensions must also be considered. By drawing on sources from a range of foundation disciplines (reliability theory, semiotics, linguistics, philosophy, logic, and ontology practice), some of the deeper and more intractable semantic issues associated with meaning exchange are exposed and explored. From this base, semantic reliability's relationship with classic reliability theory is considered, and several systems engineering techniques are identified that have potential to significantly improve semantic reliability, and hence overall system reliability.