Background -Different sets of criteria are available in veterinary dermatology for the diagnosis of canine atopic dermatitis (cAD); there are limited data to assess their reliability.Hypothesis/Objectives -The aim of this study was to assess the discriminative and predictive ability of four sets in diagnosing cAD in a population of dogs with different skin diseases.Animals -Two hundred and fifty dogs examined in the dermatology unit of a veterinary teaching hospital.Methods and materials -Dogs were diagnosed with cAD, flea infestation (FI), flea bite hypersensitivity (FBH), sarcoptic mange (SM) and other skin diseases (OD). FI, FBH and SM were pooled in one group of dermatoses mimicking cAD (MD). All nonatopic dogs were grouped as a "global population" (GP). Four criteria sets were applied to all dogs. For each set of criteria, sensitivity, specificity, positive and negative predictive values (PPV, NPV), likelihood ratios (LR) and diagnostic odds ratio (DOR) were calculated.Results -When applied to the GP, sensitivities ranged from 54 to 83%, specificities from 68 to 75% and PPVs from 35 to 43%. NPVs ranged from 87 to 94%. LRs and DORs were poor. When applied to MD, sensitivities remained unchanged, specificities and PPV were mildly higher, whereas NPVs were mildly lower and LRs and DORs were comparable.Conclusions and clinical importance -This study showed that each set of criteria had a low diagnostic reliability when used alone. The use of more discriminant criteria integrated into a thorough clinical approach excluding MDs should be considered.