Multivariate discriminant analysis of brain stem auditory evoked response component latency intervals in patients with cerebellopontine angle tumors allowed accurate detection of 90% (35 of 39) of the tumor population with response data recordable from at least one ear. Eighty-five per cent (23 of 27) of these could be detected by using information from the unaffected ear. One of 21 normal subjects was misclassified. Tumors significantly increased the I-III and III-V intervals on the side of the tumor. Increases in III-V interval latency were also observed on the unaffected side. The size of the tumor was significantly correlated with both discriminant scores derived from the analysis of the unaffected ear and delays in the III-V intervals from either the affected or the unaffected ear. These results were attributed to physiological factors and mechanical distortions of the brain stem. Regression equations derived from linear discriminant analysis for cerebellopontine angle tumors are presented and discussed in terms of their predictive validity.