A total of 555 cervical smears, originally classified as Papanicolaou classes I and II, from women in whom three years later cytologic findings consistent with moderate dysplasia, severe dysplasia, carcinoma in situ and invasive cancer were diagnosed were reviewed in order to estimate the screening error. The initial diagnosis proved to be underestimated in 17.5% of the smears. The two diagnoses correlated in 70.2% of the smears while 12.3% of the smears that contained no abnormality were judged to be inadequate for making a diagnosis, probably representing sampling errors. Quality-control measures to reduce these errors are briefly summarized.