Alternative splicing of pre-mRNA is a complex process whose outcome depends on elements reviewed in the previous chapters such as the core spliceosome units, how the core spliceosome units interact between themselves and with other splicing enhancers and repressors, primary sequence motifs, and local RNA secondary structure. Connections between RNA splicing, transcription, and other processes have also been reviewed in the previous chapters. Splicing is inherently a stochastic process: Some defective transcripts are produced and handled by mechanisms such as nonsense-mediated decay (NMD), and studies report high variability at the transcript level between cells supposedly in similar states. Nonetheless, splicing is obviously not a random process: Many determinants of splicing regulation have been identified, and experimental measurements detect highly robust and conserved splicing changes between developmental stages and tissues. These observations naturally lead to the following questions: Can we devise a method that predicts given a cellular context and the primary transcript what would be the splicing outcome? What can such a method tell us about the underlying mechanisms that govern alternative splicing?This chapter describes how these questions can be framed and addressed using machine-learning methodology. We describe how to extract putative RNA regulatory features from genomic sequence of exons and proximal introns, how to define target values based on experimental measurements of exon inclusion, how to learn a simple splicing model that optimizes the prediction the observed exon inclusion levels from the identified RNA features, and how to subsequently evaluate the learned model accuracy.