Canine mammary gland tumours (CMTs) constitute the most common cancer in female dogs and comprise approximately 50% of all canine cancers. With the advent of highthroughput technologies such as microarray and next-generation sequencing, the molecular phenotyping (classification) of various cancers has been extensively developed. The present study used a canine RNA-sequencing dataset, namely GSE119810, to classify 113 malignant CMTs and 64 matched normal samples via an unsupervised hierarchical algorithm with a view to evaluating the association between the resulting subtypes (clusters) (n = 4) and clinical and molecular characteristics. Finally, a molecular classifier was developed, and it detected 1 high-risk molecular subtype in the training dataset (GSE119810) and 2 independent validation datasets (GSE20718 and GSE22516). Our results revealed four molecular subtypes (C2-C5) in malignant CMTs. Furthermore, the normal samples constituted a distinct group in the clustering analysis. Marked significant associations were observed between the molecular subtypes (especially C5) and clinical/molecular features, including positive lymphatic invasion, high tumour grades, histopathology diagnoses, short survival and high TP53 mutation rates (ps <.05). The high-risk subtype (C5) was further characterized through the development of a cell cycle-based gene signature, which comprised 37 proliferation-related genes according to the support vector machine algorithm. This signature identified the high-risk group in both training and validation datasets (ps <.001). In the validation analysis, our potential classifier robustly predicted patients with positive lymphatic invasion, metastases and short survival. K E Y W O R D S canine mammary gland tumours, high-risk, molecular phenotyping, RNA-seq 1 | INTRODUCTION Canine mammary gland tumours (CMTs) comprise one of the most prevalent tumours in dogs and contain clinical, pathological and biological features comparable to the human counterpart. Such characteristics render CMTs an ideal candidate for animal breast cancer models. Furthermore, similar risk factors such as obesity, advancing age and environmental toxins are considered for breast cancer in both humans and dogs. 1-3 Besides traditional histopathological classifications, molecular classification (phenotyping) through gene expression profiles obtained via microarray or next-generation sequencing has been developed extensively to categorize tumoral samples into different molecular subtypes, which may correlate with or predict some clinical or biological All authors contributed equally.