Alcohol- and formalin-fixed, paraffin-embedded samples of 71 brain tumors (35 gliomas, 22 metastatic carcinomas, 8 meningiomas and 6 other tumors) were investigated by immunocytochemistry with three different monoclonal antibodies against proliferating cell nuclear antigen (PCNA)/cyclin (19A2; 19F4; PC10). PC10 was found to work best; it is applicable to both alcohol- and formalin-fixed tumor samples. PCNA labeling indices (LIs) were compared in the same tumors with LIs obtained by Ki-67 immunostaining of frozen sections and by in vitro incubation with bromodeoxyuridine (BrdUrd); in the latter preparations, BrdUrd LIs could be compared with PCNA LIs in the very same areas of serial sections. In gliomas, PCNA LIs were 0.7-80.2% (mean 31.7%), in metastases 0-76.0% (mean 47.8%), and in meningiomas 0-53.0% (mean 19.3%). In general, PCNA LIs were highly significantly correlated with Ki-67 LIs (P = 0.0002) and BrdUrd LIs (P = 0.0001). However, when tumor subgroups are considered, only gliomas show a significant correlation with Ki-67 and BrdUrd LIs. Despite this statistical correlation, PCNA expression was out of proportion to proliferation indices as determined by both other methods in almost one third of all brain tumors. Immunocytochemistry for PCNA produces a broad spectrum of staining intensity of labeled nuclei, whose number is dependent upon the sensitivity of the immunocytochemical technique used. Thus, inter-observer and inter-laboratory variabilities in PCNA LI determination may occur. Overlapping of PCNA LIs between tumor subgroups of varying malignancy further limits the informational value for the individual case.(ABSTRACT TRUNCATED AT 250 WORDS)