The dog represents an important model animal in biomedical, evolutionary and behavioral research. The current canine gene annotation has been greatly improved but still significantly lag behind in accuracy compared to those of the human and the mouse. To address this, we identified several significant issues in the current canine gene annotations. These include gene over-merging, gene name mislabeling, retrogene mis-classification, coding region mis-identification, and UTR fragmentation. We systematically addressed these issues by a novel pipeline that also utilizes our own data.