The accuracy of somatic cell counts in milk samples was investigated in four studies. First, the counts recorded by one milk buyer in one supply over six months ranged from 105,000 to 401,000 cells/ml with no apparent changes in the volume of milk consigned or the level of mastitis in the herd that would explain this wide range. Secondly, the counts in daily samples from one bulk milk supply for 28 days ranged from 84,000 to 282,000 cells/ml, again with no apparent changes in the performance of the herd to explain the wide range. Thirdly, the replicated counts recorded for one sample by three separate laboratories agreed closely; however, when a sample with a high cell count was interspersed then two of the three laboratories reported high cell counts suggestive of ‘carry-over’ in excess of the 2 per cent ‘allowable’. Finally, cell count data from three separate laboratories on samples from 21 cows for 33 days revealed problems with the misidentification of samples on the farm in 1 per cent of the samples, and misidentification and mishandling of 1 to 2·6 per cent of the samples in the laboratories. All three laboratories differentiated samples from cows with subclinical and clinical mastitis, but the mean cell count of the uninfected cows varied between the laboratories with one of them recording statistically significantly higher counts over the period.