Serum samples obtained from healthy, asymptomatic dogs in areas of Wisconsin and northern Illinois where Lyme disease is endemic or nonendemic were assayed for antibodies to Borrelia burgdorferi by enzyme-linked immunosorbent assay (ELISA), and positive results were confirmed by immunoblot assay. We found that 56.9% (562 of 1,077) of the samples were positive by ELISA and 82.0% (461 of 562) were positive by immunoblotting. A logistic regression model was developed to distinguish between nonvaccinated dogs naturally infected with B. burgdorferi from areas where the disease is endemic and dogs from areas where the disease is nonendemic that were vaccinated against Lyme disease. Of the 18 protein bands analyzed, 8 were found to be significantly different (P < 0.05) between the two groups. p93, p34, p31, and p28 occurred with increased frequency in vaccinated dogs, while p58, p37, p35, and p30 occurred more frequently in naturally infected dogs. The logistic regression equation obtained was used to determine the probability of natural infection among vaccinated dogs residing in areas where the disease is endemic. Of 125 samples, 87.2% had a very low probability of natural infection and only 2.4% were highly likely to be infected. Logistic regression is a useful method for distinguishing between vaccinated and naturally infected dogs and predicting the serological status of vaccinated dogs from areas where Lyme disease is endemic.