Host tissues affected by Leishmania infantum have differing degrees of parasitism. Previously, the use of different biological tissues to detect L. infantum DNA in dogs has provided variable results. The present study was conducted to evaluate the accuracy of molecular diagnostic testing (qPCR) in dogs from an endemic area for canine visceral leishmaniasis (CVL) by determining which tissue type provided the highest rate of parasite DNA detection. Fifty-one symptomatic dogs were tested for CVL using serological, parasitological and molecular methods. Latent class analysis (LCA) was performed for accuracy evaluation of these methods. qPCR detected parasite DNA in 100% of these animals from at least one of the following tissues: splenic and bone marrow aspirates, lymph node and skin fragments, blood and conjunctival swabs. Using latent variable as gold standard, the qPCR achieved a sensitivity of 95.8% (CI 90.4–100) in splenic aspirate; 79.2% (CI 68–90.3) in lymph nodes; 77.3% (CI 64.5–90.1) in skin; 75% (CI 63.1–86.9) in blood; 50% (CI 30–70) in bone marrow; 37.5% (CI 24.2–50.8) in left-eye; and 29.2% (CI 16.7–41.6) in right-eye conjunctival swabs. The accuracy of qPCR using splenic aspirates was further evaluated in a random larger sample (n = 800), collected from dogs during a prevalence study. The specificity achieved by qPCR was 76.7% (CI 73.7–79.6) for splenic aspirates obtained from the greater sample. The sensitivity accomplished by this technique was 95% (CI 93.5–96.5) that was higher than those obtained for the other diagnostic tests and was similar to that observed in the smaller sampling study. This confirms that the splenic aspirate is the most effective type of tissue for detecting L. infantum infection. Additionally, we demonstrated that LCA could be used to generate a suitable gold standard for comparative CVL testing.