In a study of variability in the diagnosis of epithelial abnormalities, cervical smears with abnormalities of different severity were rescreened twice by 19 observers with an interval of six months. The observers focused on grading atypicality of squamous, squamous metaplastic and endocervical columnar epithelial cells; their results were compared (1) for the two screenings to assess intraobserver variability and (2) to "review" (final) diagnoses to assess interobserver variability. When the same observer rescreened a smear, 83.3% of the diagnoses did not differ more than one grade between two screenings; however, average intraobserver variability differed considerably for individual observers. The intraobserver variability was only slightly (not significantly) influenced by the years of experience in cytopathology of the observers. Intraobserver variability proved to be an important factor in incorrect diagnoses: 49.1% of the smears with false-negative and 52.9% with false-positive diagnoses at the first rescreening were correctly assessed at the second rescreening. Of all diagnoses made at rescreening, 80.9% were in agreement with the review diagnosis. The interobserver variability also showed considerable differences between observers; however, there was a strong influence of the experience of the observer on the interobserver variability. Atypicality grading of endocervical columnar epithelium by the observers showed a low correlation with the review diagnoses. The relatively low accuracy in the evaluation of this kind of epithelial abnormality is likely to be attributable to the low incidence of abnormal changes of endocervical columnar epithelium. The results of this study point to intraobserver variability as the main cause of false diagnoses. When wrongly diagnosed, severe epithelial abnormalities are more often underestimated than completely overlooked. Apart from training in cytopathology, the establishment of laboratory protocols for multiple screening of even minor abnormalities seem to be the most effective means of reducing the number of false diagnoses.