To determine the proportion of fevers caused by leptospirosis, we obtained serum specimens and epidemiologic and clinical data from patients in Galle, Sri Lanka, March-October 2007. Immunoglobulin M ELISA was performed on paired serum specimens to diagnose acute (seroconversion or 4-fold titer rise) or past (titer without rise) leptospirosis and seroprevalence (acute). We compared (individually) the diagnostic yield of acute-phase specimens and clinical impression with paired specimens for acute leptospirosis. Of 889 patients with paired specimens, 120 had acute leptosoirosis and 241 had past leptospirosis. The sensitivity and specificity of acute-phase serum specimens were 17.5% (95% confidence interval [CI] 11.2%-25.5%) and 69.2% (95% CI 65.5%-72.7%), respectively, and of clinical impression 22.9% (95% CI 15.4%-32.0%) and 91.7% (95% CI 89.2%-93.8%), respectively. For identifying acute leptospirosis, clinical impression is insensitive, and immunoglobulin M results are more insensitive and costly. Rapid, pathogen-based tests for early diagnosis are needed.