The differentiation of malaria from other causes of fever in the absence of microscopy is notoriously difficult. Clinical predictors of malaria have been studied in an area of low and unstable transmission on the western border of Thailand. In 1527 children aged 2-15 years who were followed prospectively for 7 months, 82% (1254) had at least one febrile episode. Malaria caused 24% (301) of the first febrile episodes (Plasmodium falciparum 128, P. vivax 151, P. malariae 1, mixed infections with P. falciparum and P. vivax 21). Each malaria case was matched with the next child of similar age presenting to the dispensary with another cause of fever. Clinical symptoms or signs associated with a final diagnosis of malaria were: confirmed fever (> or = 38 degrees C) (odds ratio [OR] 1.6, 95% confidence interval [95% CI] 1.4-1.9), headache (OR 1.5, 95% CI 1.3-1.9), muscle and/or joint pain (OR 2.0, 95% CI 1.6-2.8), nausea (OR 1.7, 95% CI 1.4-2.3), clinical anaemia (OR 1.4, 95% CI 1.3-3.3), palpable spleen (OR 1.3, 95% CI 1.1-1.7), palpable liver (OR 1.4, 95% CI 1.1-2.1), absence of cough (OR 1.6, 95% CI 1.4-2.0), and absence of diarrhoea (OR 1.5, 95% CI 1.2-2.4). None of these signs alone or in combination proved a good predictor of malaria. The best diagnostic algorithms (history of fever and headache without cough, and history of fever with an oral temperature > or = 38 degrees C [sensitivity 51% for both, specificity 72 and 71%, respectively]) would result in prescription of antimalarial drugs in 28-29% of the non-malaria febrile episodes, and only 49% of the true malaria cases. Thus half of the potentially life-threatening P. falciparum infections would not be treated. Although multivariate analysis identified vomiting, confirmed fever, splenomegaly and hepatomegaly as independent risk factors for a diagnosis of falciparum malaria, use of these signs to differentiate falciparum from vivax malaria, and thus to determine antimalarial treatment, was insufficiently sensitive or specific. Malaria diagnosis should be confirmed by microscopical examination of a blood slide or the use of specific dipstick tests in areas of low transmission where highly drug-resistant P. falciparum coexists with P. vivax.