The micronucleus test in binucleated lymphocytes is a sensitive standard assay for biomonitoring, mutagenicity testing and to assess radiosensitivity of blood donors. The results vary between laboratories and scorers which led to the definition of international scoring criteria. We used these criteria in a case-control study, but nevertheless observed large differences between the seven scorers on the level of descriptive analysis. Therefore, we used the repeat measurements (267 in 98 blood donors) from this dataset (354 measurements in 185 blood donors) to analyse scoring variability in the setting of a case-control study. The variability was assessed by analysis of variance, which revealed the storage time of the blood samples, the blood donors including their disease status, and the scorers as sources of variation in the entire dataset. In addition, the coefficient of variation (CV) of the measurements was determined (overall: CV = 24.3%). After stepwise removal of biological and experimental variation by normalizations, the CV dropped to 6.8% on average, which may reflect the 'pure counting error'. The scorer-specific CVs were between 5.5 and 9.5%. The differences between the scorers suggested by the raw data were neither related to the scorer-specific CV nor to their experience. Instead, we observed a general decline of the micronuclei frequencies towards the end of the study for all scorers. This could not be related to a change in experimental conditions or in the defined scoring criteria. An explanation could be an unintended and unrecognized change of scoring criteria. Since the change in the results did not occur in automated counting we suggest to use either reference slides in longer-lasting studies or automated counting by image analysis.