(1) 419 samples were counted in each of ten stations (laboratories) in order to assess the accuracy of the plate-count method under the conditions existing in practice.

(2) When, in each count, that plate which would ordinarily be used in practice is selected for the final count, it is found that the range which would include 95 per cent, of counts in the same sample is as wide as from one-ninth of the mean to nine times the mean when inter-station differences are taken into account. For 80 per cent, of counts in the same sample the corresponding range would be from 1/(4·1) of the mean to 4·1 times the mean.

(3) Systematic differences between stations under these conditions account for between 3 and 4 per cent, of the total experimental variation.

(4) When thirty-one complete samples in which all counts were made at a dilution of 1/10 were statistically analysed separately it was found that the range which would include 95 per cent, of counts in the same sample was from 1/(3·7) of the mean to 3·7 times the mean. For 80 per cent, of counts in the same sample the corresponding range would be from 1/(2·3) of the mean to 2·3 times the mean. There were no systematic station differences.

(5) When all plates are counted at the same dilution the error is therefore much reduced and this result suggests that the systematic station differences exhibited by the 419 samples as well as some of the random errors occurring in them is due to the combining of plate counts made at different dilutions.

In effect this means that owing to the different conditions encountered during transit comparable figures for any sample could not be expected.