Histological grading of canine mammary carcinomas (CMCs) has been performed using an adaptation of the human Nottingham method. The histological grade could be a prognostic factor in CMC; however, no data are available concerning interobserver variability in grading. In this study we analyzed the interobserver reproducibility between three observers when assigning individual parameter scores and grade to 46 CMCs. The influence of tumour size and vascular invasion and/or lymph node metastases on the odds of grading disagreement was also evaluated. The mean kappa values were 0.71, 0.51, 0.69 and 0.70 for tubule formation, nuclear pleomorphism, mitotic counts and grade, respectively. There was moderate to good agreement in scoring parameters and tumour grading, with nuclear pleomorphism being least reproducible. These findings are similar to those of human studies. The odds of grading disagreement increased with tumour size, but decreased with the presence of vascular invasion and/or lymph node metastases. Individual scoring differences were moderated by reaching a consensus between two observers.