The elimination of seropositive dogs in Brazil has been used to control zoonotic visceral leishmaniasis but with little success. To elucidate the reasons for this, the infectiousness of 50 sentinel dogs exposed to natural Leishmania chagasi infection was assessed through time by xenodiagnosis with the sandfly vector, Lutzomyia longipalpis. Eighteen (43%) of 42 infected dogs became infectious after a median of 333 days in the field (105 days after seroconversion). Seven highly infectious dogs (17%) accounted for >80% of sandfly infections. There were positive correlations between infectiousness and anti-Leishmania immunoglobulin G, parasite detection by polymerase chain reaction, and clinical disease (logistic regression, r2=0.08-0.18). The sensitivity of enzyme-linked immunosorbent assay to detect currently infectious dogs was high (96%) but lower in the latent period (<63%), and specificity was low (24%). Mathematical modeling suggests that culling programs fail because of high incidence of infection and infectiousness, the insensitivity of the diagnostic test to detect infectious dogs, and time delays between diagnosis and culling.