Summary By means of a mathematical score previoulsy generated by discriminant analysis on 90 lung cancer patients, a new and larger group of 261 subjects [209 with non-small-cell lung cancer (NSCLC) and 52 with small-cell lung cancer (SCLC)] was analysed to confirm the ability of the method to distinguish between these two types of cancers. The score, which included the serum neuron-specific enolase (NSE) and CYFRA-21.1 levels, permitted correct classification of 93% of the patients. When the misclassifications were analysed in detail, the most frequent errors were associated with limited disease SCLC with low NSE levels and with advanced NSCLC with high NSE levels. This demonstrates the importance of the marker in correctly categorizing patients.