This thesis addresses the problem of designing roof support systems in coal mines. When designing the roof support, it is necessary to account for the uncertainties that are inherently exist within the rock mass and support elements. The performance of a support system is affected by these uncertainties, which are not taken into account in the current design methodologies used in South Africa. This study sets out to develop a method which takes all uncertainties into account and quantitatively provides a risk-based design. Despite the fact that the roof bolting is probably one of the most researched aspects of coal mine ground control, falls of ground still remain the single major cause of fatalities and injuries in South African collieries. Mainly five different support design methodologies have been used; namely, analytical modelling, numerical modelling, physical modelling, design based on geotechnical rating systems and field testing. As part of this study, it is shown that there are many elements of a support system that can impact the support and roof behaviour in a coal mine and the characteristics of these elements as well as the interaction between them is complex and can vary significantly within a short distance. These variations account for uncertainties in coal mine roof support and they are usually not taken into account in the above design methodologies resulting in falls of ground and/or over design of support systems. The roof and support behaviour were monitored at 29 sites at five collieries. It is found that there was no evidence of a dramatic increase in the stable elevations as experienced in some overseas collieries. A roadway widening experiment was carried out to establish the critical roof displacements. The maximum width attained was 12 m at which stage 5 mm displacement was measured. During the monitoring period no roof falls occurred at any of the 29 sites and road widening experiment site, even where 12 mm displacements were measured. The in situ monitoring programme was continued in additional 26 monitoring stations in 13 sites with the aim of establishing the effect of unsupported cut-out distance on roof and support performances. The results showed that the lithological composition of the roof strata plays a major role in the amount of deflection that was recorded. Bedding separation was seen to occur at the contacts between different strata types. It is concluded that the roof behaved like a set of composite beams with different characteristics. It is also found that the amounts of deflection corresponded with the deflection that would be expected from gravity loaded beams. During this monitoring programme variable nature of roof and support systems are also demonstrated. As many mines use different geotechnical rating systems, an evaluation of the currently used classification techniques were conducted to determine their effectiveness in design of roof support strategies. It is found that currently used systems cannot quantitatively determine the required support system in a given geotechnical environment. Impact splitting tests are found to be the appropriate system for South African conditions. It is however concluded that the roof lithology, stress regime and roof characteristics can change within meters in a production section. Therefore, in order to predict these changing conditions many boreholes are required for a section, which would be costly and time consuming. An in-depth study into the roof support elements was conducted for the purpose of obtaining an understanding of the fundamental mechanisms of roof support systems and developing guidelines for their improvement. All of the currently available roof bolt support elements and related machinery were evaluated using in situ short encapsulated pull tests. The results showed that, on average, bond strengths obtained from the roof bolts supplied by different manufacturers can vary as much as 28 per cent. The test results conducted on different resins showed that the strength of resin currently being used in South Africa is adequate. Differences between commonly used bit types were established. It is concluded that the 2-prong bit outperforms the spade bit in sandstone and shale rock types. In addition, the effect of hole annulus was also investigated as part of this study. The results show that an annulus between 2.5 mm to 3.8 mm resulted in the most effective bond strengths. The effect of wet and dry drilling was noted. It is found that bond strengths and overall support stiffnesses are greater with the use of the wet drilling in all resin types. The results from the tests in different rock types highlighted the very distinct differences between bolt system performances. Quality control procedures for compliance with the design, support elements and quality of installation are presented. Recommendations for improving the quality control measures and for developing testing procedures for bolt system components, installation quality and resin performance are provided. Finally, a roof support design methodology that takes into account all natural variations exist within the rock mass and the mining process has been developed and presented. This was achieved by adapting a probabilistic design approach using the well established stochastic modelling technique. This methodology enables rock engineers to design roof support systems with greater confidence and should result in safer and economic extraction of coal reserves.