Data are presented for 2382 children investigated for fever in a Malaysian hospital between 1984 and 1987 when Widal tests and blood cultures were a routine part of every fever screen. There were 145 children who were culture positive (TYP-CP) for Salmonella typhi, while 166 were culture negative but were diagnosed as having typhoid (TYP-CN). Analyses of the sensitivity and specificity of combinations of initial Widal titres in predicting a positive S. typhi culture in a febrile child (culture positive vs the rest) showed the best model to be an O- and/or H-titre of > or = 1 in 40 (sensitivity 89%; specificity 89%). While the negative predictive value of the model was high (99.2%) the positive predictive value remained below 50% even for very high titres of O and H (> 1 in 640), at which point the specificity was 98.5%, supporting the clinical view that a high proportion of the TYP-CN patients really were typhoid but were missed by culture. The TYP-CN patients showed a very similar clinical and age profile to TYP-CP patients. The length of history of fever did not affect the initial Widal titre in culture positive cases. The Widal test in children remains a sensitive and specific 'fever screen' for typhoid although it will not identify all cases. In children, lower cut-off points for O- and H-titres should be used than are generally recommended.