The diagnosis of childhood illness by maternal health interview surveys is widely used to estimate the prevalence of childhood morbidity in developing countries. To determine the validity of interview-based diagnoses, and to define simple, sensitive and specific diagnostic algorithms, we compared symptoms and signs reported by mothers during structured interviews with physicians' diagnoses for 271 children on the Philippine island of Cebu. The 271 children had 318 physician diagnosed illnesses: 105 acute lower respiratory infections (ALRI), 121 diarrhoeas, 36 measles, 50 upper respiratory infections (URTI), 5 roseola infantums and one milaria rubria. An algorithm for measles (age greater than or equal to 120 days, rash and fever greater than or equal to 3 days and fading of rash) had a sensitivity and specificity of 94%. For ALRI an algorithm of cough, dysponea and fever had a sensitivity of 82%, but specificity was lower in comparison with URTI (58%) than with children who had no respiratory illness (79%). Inclusion of signs of respiratory distress (flaring of nostrils, intercostal retraction) raised the specificity to 83-84%, but reduced sensitivity to 68%. Diagnosis of diarrhoea based on frequent loose or liquid stools had a sensitivity of 95-97% and specificity of 80% in children with or without concomitant non-diarrhoeal illnesses. Addition of questions on numbers of stools (greater than or equal to 6 per day), and no signs of dehydration increased specificity to 95% but reduced the sensitivity to 84-86%. However, specific signs of dehydration were not well reported by the mothers.(ABSTRACT TRUNCATED AT 250 WORDS)