One of the main biological goals of recent years is to determine the genes in the human genome that cause disease. Recent technological advances have realised genome-wide association studies, which have uncovered numerous genetic regions implicated with human diseases. The current approach to analysing data from these studies is based on testing association at single SNPs but this is widely accepted as underpowered to detect rare and poorly tagged variants. In this thesis we propose several novel approaches to analysing large-scale association data, which aim to improve upon the power offered by traditional approaches. We combine an established imputation framework with a sophisticated disease model that allows for multiple disease causing mutations at a single locus. To evaluate our methods, we have developed a fast and realistic method to simulate association data conditional on population genetic data. The simulation results show that our methods remain powerful even if the causal variant is not well tagged, there are haplotypic effects or there is allelic heterogeneity. Our methods are further validated by the analysis of the recent WTCCC genome-wide association data, where we have detected confirmed disease loci, known regions of allelic heterogeneity and new signals of association. One of our methods also has the facility to identify the high risk haplotype backgrounds that harbour the disease alleles, and therefore can be used for fine-mapping. We believe that the incorporation of our methods into future association studies will help progress the understanding genetic diseases.