Abstract Motivation Recent advances in high-throughput RNA-Seq technologies allow to produce massive datasets. When a study focuses only on a handful of genes, most reads are not relevant and degrade the performance of the tools used to analyze the data. Removing irrelevant reads from the input dataset leads to improved efficiency without compromising the results of the study. Results We introduce a novel computational problem, called gene assignment and we propose an efficient alignment-free approach to solve it. Given an RNA-Seq sample and a panel of genes, a gene assignment consists in extracting from the sample, the reads that most probably were sequenced from those genes. The problem becomes more complicated when the sample exhibits evidence of novel alternative splicing events. We implemented our approach in a tool called Shark and assessed its effectiveness in speeding up differential splicing analysis pipelines. This evaluation shows that Shark is able to significantly improve the performance of RNA-Seq analysis tools without having any impact on the final results. Availability and implementation The tool is distributed as a stand-alone module and the software is freely available at https://github.com/AlgoLab/shark. Supplementary information Supplementary data are available at Bioinformatics online.