The objectives of this study were to compare the sensitivities and reliabilities of different PCR methods for the diagnosis and epidemiological study of canine visceral leishmaniasis (CVL) using dog blood. We chose to work with peripheral blood, as this type of sampling is noninvasive, straightforward, and easy to repeat. Six PCR methods were compared: three primer pairs target genomic DNA, and the other three target kinetoplast (mitochondrial) DNA. Sensitivity, specificity, reproducibility, and ease of interpretation without hybridization were evaluated for each method. The assessment was first performed using artificial samples. All methods could detect less than one parasite per reaction tube. However, the sensitivities varied among the different methods by a factor of 500 on purified cultivated parasites and by a factor of 10,000 on seeded dog blood samples (i.e., from 10 to 10(-3) parasite per ml of blood for the latter). Only four methods were found sufficiently reliable for the diagnosis of CVL. They were tested on 37 dogs living in an area of endemicity and grouped according to clinical status and specific serology. Only the two methods targeting kinetoplast DNA (K13A-K13B and RV1-RV2) could detect the parasite in 100% of symptomatic infected dogs. Similarly, all seropositive dogs were found PCR positive by these methods versus 62% by the genomic-DNA-based methods. Finally, these kinetoplast-based methods proved clearly superior to the others in the detection of Leishmania in asymptomatic dogs. Our data allow the discussion of the advantages and drawbacks of highly sensitive versus moderately sensitive PCR methods in diagnosis and prevalence studies of CVL.