In numerous species, low birth weight is a risk factor for neonatal mortality. In the canine species, definition of a low birth weight is complex due to the huge interbreed variability in size. To identify puppies at higher risk of neonatal death, data from 6,694 puppies were analysed. The data were collected from 75 French breeding kennels, examining 27 breeds and totaling 1,202 litters of puppies. Generalised linear mixed models allowed to identify birth weight, birth weight heterogeneity within the litter, and size of the breeding kennel as significant risk factors for neonatal mortality. Receiver Operating Characteristics (ROC) and classification and regression tree (CART) analyses were combined to define breed specific thresholds for birth weight allowing the identification of puppies at higher risk of neonatal mortality. Due to differences in birth weights between breeds, including when belonging to the same breed size, analyses were conducted at the breed level. First, ROC analysis thresholds were successfully established for 12 breeds (area under the ROC ≥ 0.70; sensitivity ≥ 75%; specificity: 45–68%) and they ranged from 162 g in the Maltese to 480 g in the Bernese Mountain dog. Secondly, CART analysis thresholds from 22 breeds ranged from 105 g in the Maltese and 436 g in the Boxer. Puppies were grouped into three categories according to birth weight: low, moderate and high risk of neonatal mortality (higher than the ROC threshold, between ROC and CART thresholds, and lower than the CART threshold respectively). In the current study, 44% of the puppies were classified as at moderate risk and 5.3% for a high risk of neonatal mortality. Thresholds defined by CART analysis (and not ROC analysis) were used to define low birth weight puppies and were sometimes quite different between breeds with similar birth weight distributions suggesting a variable relationship between birth weight reduction and neonatal death. These results allow the identification of puppies at an increased risk of neonatal death, thus requiring specific nursing to improve their chances of survival. With these high risk puppies identified, both animal welfare and kennel productivity is predicted to improve.