In a preliminary investigation into the experimental error of the plate count examination of butter, 154 boxes of choicest quality butter were examined by plating in triplicate, in a dilution of 1/500, each of three 1 g. samples per box.

X2 tests in both this and a later series indicated that the means of the triplicate counts in respect of plates averaging less than 300 colonies gave, in the main, reasonably satisfactory estimates of the bacterial content of the samples. In two of the tests, however, excessive within-sample variability was shown by abnormally high values obtained from the summation of x2. This is accepted as an indication that the plate count was not always a reliable index of the sample population.

The between-sample variability in this series has been shown by transforming the counts to logarithms and calculating the coefficient of variation in respect of each box. In the distribution of the coefficients approximately 50% were found to lie on either side of a 4·0% level, while 10% gave values higher than 14·0%.

In a further inquiry twelve boxes of butter were selected for quality and uniformity and data were secured from the examination of seven 1 oz. samples selected at random from each box.

Treatment of the results by the method of analysis of variance, using a square root transformation, indicated that the estimates of within-sample variance obtained could be regarded as estimates of a common variance. Highly significant differences were found, however, among the between-sample mean squares, and the variability was found to be excessive in six of the twelve boxes examined.

No significant general correlation was found in this series between bacterial counts and either salt or moisture determinations.

The application of these results to research problems and to routine tests involving the estimation of bacterial numbers in butter is discussed.