Introduction Dengue is prevalent in many tropical and sub-tropical regions. The clinical diagnosis of dengue is still complex, and not much data are available. This work aimed at assessing the diagnostic accuracy of the tourniquet test in patients with suspected dengue infection and its positivity in different classifications of this disease as reported to the Information System for Notifiable Disease in Belo Horizonte, State of Minas Gerais, Brazil between 2001 and 2006. Methods Cross-section analysis of the diagnostic accuracy of the tourniquet test for dengue, using IgM-anti-DENV ELISA as a gold standard. Results We selected 9,836 suspected cases, of which 41.1% were confirmed to be dengue. Classic dengue was present in 95.8%, dengue with complications in 2.5% and dengue hemorrhagic fever in 1.7%. The tourniquet test was positive in 16.9% of classic dengue cases, 61.7% of dengue cases with complications and 82.9% of cases of dengue hemorrhagic fever. The sensitivity and specificity of the tourniquet test were 19.1% and 86.4%, respectively. Conclusions A positive tourniquet test can be a valuable tool to support diagnosis of dengue where laboratory tests are not available. However, the absence of a positive test should not be read as the absence of infection. In addition, the tourniquet test was demonstrated to be an indicator of dengue severity.