The mining industry in South Africa contributes significantly to the national economy. Despite stringent safety legislation, mining accidents cause numerous fatalities and injuries. Inadequate or insufficient training is often cited as a root cause of accidents. Conventional class-based safety training has not reduced the incidence of accidents significantly. By contrast, virtual reality training tools can provide simulated exposure to real-world working conditions without the associated risks. This study describes the application of design-based research (DBR) in the design and development of two desktop virtual reality (VR) systems for safety training in the South African mining industry. The results of a usability context analysis were applied in the design of a VR prototype on generic hazards recognition and rectification, which was used and evaluated at South Africa‘s largest platinum mine site. A case study was conducted to investigate the causes and occurrences of falls of ground, which resulted in the design and development of a second VR prototype focusing on identifying and addressing underground geological conditions. DBR was also used in the generation of an evaluation framework for evaluating VR training systems, namely the Desktop VR Evaluation Framework (DEVREF), which is the major deliverable of the research. DEVREF can make a major contribution to the domain of e-training in mines and is transferable and customisable beyond its initial application. The process flow of the research thus moved beyond merely providing a solution to a complex real-world problem and became a classic DBR study with dual outcomes, namely a practical real-world solution in the form of two VR training systems and a theoretical contribution in the form of the DEVREF evaluation framework. DEVREF evaluates the design of desktop VR training systems in the categories of instructional design, usability, VR systems design, and context-specific criteria for mining. The use of DEVREF is demonstrated by reporting the application of its criteria in evaluating the two VR training systems. Heuristic evaluation, end-user surveys, and interviews were used as evaluation methods. A third contribution is methodological, in that this work proposes a new DBR process model and an interaction design lifecycle model suitable for VR training systems. / Computing / D. Phil. (Information Systems)