We know little about how isoform choice is regulated in individual cells for most spliced genes. In theory, single-cell RNA-sequencing (scRNA-seq) could enable us to investigate isoform choice at cellular resolution. Therefore, scRNA-seq could give insight into the fundamental molecular biology process of how alternative splicing is regulated within cells. However, scRNA-seq is a relatively new technology, and at the start of my PhD it was not clear whether existing bioinformatics approaches would enable accurate splicing analyses. In my PhD I consider what the limitations are when attempting to study alternative splicing using scRNA-seq and what can be done to overcome them. Alternative splicing is commonly analysed using bulk RNA sequencing (bulk RNA-seq) data with isoform quantification software. It was not clear whether isoform quantification software designed for bulk RNA-seq would perform well when run on scRNA-seq data. To address this, I performed a simulation-based benchmark of isoform quantification software developed for bulk RNA-seq when run on scRNA-seq. I made two important findings. Firstly, I found that isoform quantification software performs poorly when run on Drop-seq data, but performs better when run on scRNA-seq data generated using full-length transcript protocols (eg. SMART-seq and SMART-seq2). Secondly, I found that for the most part, isoform quantification software performs almost as well when run on full-length scRNA-seq as it does when run on bulk RNA-seq. Based on these findings, I concluded that software tools to accurately quantify the reads from full-length scRNA-seq experiments exist, theoretically enabling alternative splicing to be analysed using scRNA-seq. Encouraged by this result, I embarked on a series of experiments designed to answer questions such as ‘How many isoforms does a gene typically produce per cell?’. This is a key basic biology question that could in theory be answered using scRNA-seq. Unfortunately, I found that the results of these experiments were largely impossible to interpret because I was unable to distinguish between biological signal and technical noise. I realised that without a solid understanding of the technical noise and confounding factors associated with scRNA-seq, distinguishing biological signal from technical noise would be challenging and might not be possible. To address this, I embarked on a second simulation-based study, this time investigating the impact of technical noise on our ability to study alternative splicing using scRNA-seq. I simulated four situations: a situation where every gene expressed one isoform per cell, a situation where all genes expressed two isoforms per cell, a situation where all genes expressed three isoforms per cell and a situation where all genes expressed four isoforms per cell. Importantly, I explicitly simulated isoform choice, dropouts and quantification errors. The results of the four simulated situations were not trivial to distinguish from each other, raising concerns about the feasibility of resolving the more complex splicing patterns that probably exist in reality using scRNA-seq data. I concluded that attempts to study alternative splicing using scRNA-seq are currently substantially confounded by a high rate of dropouts and a lack of understanding about the mechanism of isoform choice. Importantly, improvements to isoform quantification software accuracy alone were insufficient to correct for confounding effects caused by dropouts. I propose that to enable accurate alternative splicing analyses using scRNA-seq, further research into accurately modelling dropouts is required, or alternatively, scRNA-seq technologies should be improved to increase their capture efficiency. Additionally, research into how isoform choice is regulated at a cellular level is necessary to enable accurate analyses. Overall, I find that it is not currently possible to accurately perform alternative splicing analyses using scRNA-seq. However, I am optimistic that with further research, it may become possible in the future.