The zebrafish is ideal for studying embryogenesis and is increasingly applied to model human disease. In these contexts, RNA-sequencing (RNA-seq) provides mechanistic insights by identifying transcriptome changes between experimental conditions. Application of RNA-seq relies on accurate transcript annotation for a genome of interest. Here, we find discrepancies in analysis from RNA-seq datasets quantified using Ensembl and RefSeq zebrafish annotations. These issues were due, in part, to variably annotated 3' untranslated regions and thousands of gene models missing from each annotation. Since these discrepancies could compromise downstream analyses and biological reproducibility, we built a more comprehensive zebrafish transcriptome annotation that addresses these deficiencies. Our annotation improves detection of cell type-specific genes in both bulk and single cell RNA-seq datasets, where it also improves resolution of cell clustering. Thus, we demonstrate that our new transcriptome annotation can outperform existing annotations, providing an important resource for zebrafish researchers.