Despite advances in sequencing technologies, many patients with genetic epilepsies still face diagnostic challenges. To uncover a genetic cause in these patients, screening will need to extend to regions not covered by conventional analyses. Variants identified in introns, the non-coding sequences in genes, are often disregarded as they are frequently occurring and difficult to interpret. However, recent studies have identified deep-intronic variants that cause disease by disrupting the process of gene splicing. An increasing number of computational tools that classify the impact of deep-intronic variants have been developed, but they are not yet utilised in mainstream analysis pipelines as their performance is not well characterised. To address this, we have compiled a truth dataset of known pathogenic and benign deep-intronic variants to evaluate the performance of several variant prediction tools. Our results indicate that the machine-learning tool Pangolin has the highest sensitivity in these regions as it correctly classified 84% of known pathogenic variants. We then applied a selection of in-silico tools to the sequencing data of unsolved epilepsy patients to identify and prioritise candidate variants. We are evaluating the contribution of deep-intronic splice variants to genetic epilepsies as it is believed that they are an underappreciated disease mechanism.