Brain-computer interface (BCI) is a fast growing field of research that utilises non-invasive Electroencephalography (EEG) to enable control and communication for both patients and healthy users. The aim of this thesis is to provide a comprehensive improvement step for BCI technology. This will be focussed around three key values; robustness, human-centred development, and innovation. Firstly, BCI robustness is enhanced by critically analysing the literature and, for the first time, justifying the variations in classification accuracies. This was achieved by conducting a literature survey of emotion recognition studies and concluding, by re-implementation, that the cause of these variations is the use of non-nested cross-validation when testing a neural network (NN) classifier. Secondly, human-centred development is conducted by examining user experience and reducing user fatigue associated with flickering lights in visual BCI. This was achieved by design and implementation of a non-flickering visual BCI paradigm that enables communication by looking at pictures of faces and non-face. The brain behaviour of visual perception was then used classify choices. The convenience of non-flickering picture was validated by 88% of participants in a survey, and the paradigm was tested using a commercial-grade EEG. Finally, innovation is created by a novel application of emotion recognition using EEG to classify violin types associated with different sound qualities. The aim of the experiment was to provide insights into brain activity during the perception process. This was achieved by conducting EEG listening tests of specifically-comprised melodies to express various emotions, played by the two violins. The two violins were classified with a statically significant accuracy of 61%. This thesis provides contributions in various ways, such as literature reviews, classifier optimisation, electronic design and implementation, survey conducting, EEG data collection, and study replications.