The ability to automatically classify human actions and activities using vi- sual sensors or by analysing body worn sensor data has been an active re- search area for many years. Only recently with advancements in both fields and the ubiquitous nature of low cost sensors in our everyday lives has auto- matic human action recognition become a reality. While traditional sports coaching systems rely on manual indexing of events from a single modality, such as visual or inertial sensors, this thesis investigates the possibility of cap- turing and automatically indexing events from multimodal sensor streams. In this work, we detail a novel approach to infer human actions by fusing multimodal sensors to improve recognition accuracy. State of the art visual action recognition approaches are also investigated. Firstly we apply these action recognition detectors to basic human actions in a non-sporting con- text. We then perform action recognition to infer tennis events in a tennis court instrumented with cameras and inertial sensing infrastructure. The system proposed in this thesis can use either visual or inertial sensors to au- tomatically recognise the main tennis events during play. A complete event retrieval system is also presented to allow coaches to build advanced queries, which existing sports coaching solutions cannot facilitate, without an inordi- nate amount of manual indexing. The event retrieval interface is evaluated against a leading commercial sports coaching tool in terms of both usability and efficiency.