For centuries, the use of glass in buildings was essentially restricted to functions such as windows and glazing. Over the last decades, continuous improvements in production and refining technologies have enabled glass elements to carry more substantial superimposed loads and therefore achieve a more structural role. The structural design of such elements, however, remains problematic. Current widely used design methods suffer from notable shortcomings. They are, for instance, not applicable to general conditions, but are limited to special cases like rectangular plates, uniform lateral loads, constant loads, time-independent stress distributions and the like. Some model parameters combine several physical aspects, so that they depend on the experimental setup used for their determination. The condition of the glass surface is not represented by user-modifiable parameters, but is embedded implicitly. The design methods contain inconsistencies and give unrealistic results for special cases. Different models yield differing results and several researchers have expressed fundamental doubts about the suitability and correctness of common glass design methods. The lack of confidence in "advanced" glass models and the absence of a generally agreed design method result in frequent time-consuming and expensive laboratory testing and in inadequately designed structural glass elements. The present thesis endeavours to improve this situation. After outlining the fundamental aspects of the use of glass as a building material, an analysis of present knowledge was conducted in order to provide a focus for subsequent investigations. Then a lifetime prediction model for structural glass elements was established based on fracture mechanics and the theory of probability. Aiming at consistency, flexibility, and a wide field of application, this model offers significant advantages over currently used models. It contains, for instance, no simplifying hypotheses that would restrict its applicability to special cases and it offers great flexibility with regard to the representation of the surface condition. In a next step, possible simplifications of the model and the availability of the model's required input data were discussed. In addition to the analysis of existing data, laboratory tests were performed and testing procedures improved in order to provide more reliable and accurate model input. In the last part of the work, recommendations for structural design and testing were developed. They include, among other things, the following: Glass elements that are permanently protected from damage can be designed by extrapolation of experimental data obtained from as-received or homogeneously damaged specimens. The design of exposed glass elements whose surfaces may be damaged during their service lives (for example, because of accidental impact or vandalism), however, should be based on a realistic estimation of the potential damage (design flaw). Appropriate predictive models and testing procedures are proposed in this thesis. If substantial surface damage has to be considered, the inherent strength contributes little to the resistance of heat treated glass. Therefore, quality control measures that allow the use of a high design value for the residual surface stress are very efficient in terms of economical material use. Results from laboratory testing at ambient conditions represent a combination of surface condition and crack growth. The strong stress rate dependence of the latter, which was demonstrated in this thesis, diminishes the accuracy and reliability of the results. The problem can be addressed by the near-inert testing procedure that was developed and used in this thesis. The application of the proposed models and recommendations in research and practice is facilitated by GlassTools, the computer software that was developed as part of this thesis.