EBSseq is an empirical Bayes hierarchical model designed to identify differentially expressed (DE) isoforms in RNA-seq experiments. The model addresses the challenge of estimating isoform expression, which is more complex than gene-level estimation due to the presence of multiple isoforms sharing a parent gene. Traditional count-based methods for gene-level inference are not suitable for isoform-level inference because they do not account for the varying uncertainty in isoform expression estimates. EBSseq models isoform expression directly, partitioning estimation uncertainty into groups defined by isoform complexity, and thus accommodating the systematic differences in variability among these groups. The model uses a mixture distribution to account for both expected and observed counts, and posterior probabilities are obtained via Bayes' rule. Simulation studies and case studies demonstrate that EBSseq has improved power and performance for identifying DE isoforms compared to other methods, including Cuffdiff2 and BitSeq. Additionally, EBSseq shows slightly increased power for identifying DE genes, with well-controlled false discovery rates (FDRs). The R package implementing EBSseq is available online, and it can handle various estimation methods for isoform expression.EBSseq is an empirical Bayes hierarchical model designed to identify differentially expressed (DE) isoforms in RNA-seq experiments. The model addresses the challenge of estimating isoform expression, which is more complex than gene-level estimation due to the presence of multiple isoforms sharing a parent gene. Traditional count-based methods for gene-level inference are not suitable for isoform-level inference because they do not account for the varying uncertainty in isoform expression estimates. EBSseq models isoform expression directly, partitioning estimation uncertainty into groups defined by isoform complexity, and thus accommodating the systematic differences in variability among these groups. The model uses a mixture distribution to account for both expected and observed counts, and posterior probabilities are obtained via Bayes' rule. Simulation studies and case studies demonstrate that EBSseq has improved power and performance for identifying DE isoforms compared to other methods, including Cuffdiff2 and BitSeq. Additionally, EBSseq shows slightly increased power for identifying DE genes, with well-controlled false discovery rates (FDRs). The R package implementing EBSseq is available online, and it can handle various estimation methods for isoform expression.