This paper describes a study of the development and implementation of a benchmarking methodology within a coaching community. It asks: how robust and valid was the process used to identify the key coaching skills and their specific behavioural indicators? It finds that the initial identification of core coach skills was driven by a literature review and by modelling based on Neurolinguistic Programming (NLP). The ongoing collaborative testing and refinement of the benchmarking meets established criteria of valid action research. This approach to assessing coach competencies is based on making fine-grained behavioural distinctions, uses direct observation and is used both for training new coaches and maintaining practising coaches’ skills. Further research is needed with regard to inter-rater reliability, outcomes and coach supervision.