Clinical records and biomedical literature, which have grown exponentially in recent years, are important for clinicians to provide personalised treatments and for individual patients to understand their health conditions well. However, essential information is often expressed in natural language. Those expressions in biomedical and clinical domains are distinct, making their processing a daunting task for automated systems. This thesis is the first comprehensive study focussing on concept extraction and multi-level normalisation across biomedical and clinical domains. In this research, we describe our work on 1) developing machine learning-based methods to recognise phenotypic concepts from biomedical and clinical articles; 2) analysing the semantic, syntactic, morphological and lexical characteristics of concepts in these two heterogeneous domains and based on this analysis, 3) proposing a normalisation method for linking phenotypic mentions from clinical records and biomedical literature to terminology standards.