This study assessed the diagnostic accuracy of formal diagnostic criteria for canine compulsive disorder (canine CD). Canine CD is a syndrome of abnormal behaviors that are believed to result from conflict or frustration. Differential diagnoses include normal conflict behavior and learned behavior. In studies of canine CD, confidence in the diagnosis comes with knowing the accuracy of the diagnostic method. This accuracy may be quantified as the chance-corrected agreement between the diagnostic method and a 'gold standard' diagnostic test. The present study examined the agreement between diagnoses of canine CD made by an expert (the 'gold standard') and by using formal diagnostic criteria. The owners of 84 dogs suspected of having CD received 2 telephone interviews. The first utilized a detailed, pre-tested questionnaire; a dog was then diagnosed with CD if the behavioral history met 7 diagnostic criteria. The second interview was given by a behavioral expert whose diagnosis was based on personal experience. The interviewers were blind to each other's diagnoses. The chance-corrected agreement between diagnoses was minimal (kappa = 0.02) and disagreement was associated with 3 of the formal criteria: a history of conflict or frustration, an increase in the number of contexts that elicit the behavior, and an increase in the daily frequency of the behavior. Reasons for the disagreement include the order of the interviews, response biases, the setting of the interviews, and, possibly, the diversity of the behaviors associated with canine CD. To the authors' knowledge, this type of study is the first in clinical ethology to address validation of the diagnostic method. The results indicate 3 developmental aspects of canine CD that should be examined in future work.