The precision of a grading system based upon average score of 8 items for histological evaluation of malignancy of epidermoid carcinomas of the larynx was studied by letting 6 pathologists examine biopsies from 22 new cases. Exploratory statistical tools as well as a statistical framework (variance component models) for separating and evaluating the noise components are suggested. The order of magnitude of the resulting overall measurement uncertainty allowed discrimination between patients, though if a patient is graded by only one pathologist, only results at the extremes of the scale of malignancy should be trusted as possible therapeutical guidance. Based upon an analysis of the basic structure of the grading system the single noise components making up the measurement uncertainty were determined. Of greatest importance was noise created by disagreement between the pathologists in their ranking of the patients on the scale of malignancy. Furthermore the pathologists used the individual items differently, and certain items seemed to conform little to the scoring system. An attempt to minimize these two last components was performed; however neither calibration of the pathologists nor omission of the deviating items improved the precision significantly.