Sources of variability in quantitating proliferative cell nuclear antigen (PCNA) by image analysis were evaluated in paraffin sections of 18 ovarian tumors of low malignant potential (LMP) and grade 1 (G1) carcinomas. The correlation coefficient of reliability (R) was calculated to determine how reliable a single observation was for representing a "true" tumor value. Reliability of 61% was obtained when interobserver and intraobserver variability were assessed. Threshold settings for positive nuclear and antibody signals minimally affected the overall reliability. The reliability of a single block of tumor for representing true tumor value was R = 0.61. These sources of variability render this technique impractical for evaluating proliferative characteristics in LMP and G1 common epithelial ovarian tumors with PCNA staining.