Employing 421 urine samples taken randomly from normal infants and 294 samples from infants with neuroblastoma, we investigated the application of values of vanillylmandelic acid (VMA), homovanillic acid (HVA) and square root of VMA2 + HVA2 to discriminate between patients and normal infants. The establishment of an HVA cut-off level is virtually impossible because of its low sensitivity. In contrast, VMA sensitivity is high, and that of square root of VMA2 + HVA2 is very similar to that of VMA although it includes HVA. Although VMA appears to be able to discriminate well, discrimination employing VMA alone would be dangerous because cases with normal VMA and abnormal HVA would be missed, especially in the first screening for which a mechanical discrimination is performed. Overlooking such cases can be avoided by the adoption of the function square root of VMA2 + HVA2. The present study suggests that the combined use of variables can be more effective than using them separately.