COLONIES ALLOWABLE ON AGAR PLATES 329 



tween 0.5 and 16.5 with a single case where the average of the 

 two plates was 44. 



If we arbitrarily assume that plates giving a count more 

 than 2500 above or below the average fail to check with the 

 accepted count, we find that the averages of all but three of the 

 100 pairs of 1: 100 plates check with the accepted count while 

 there are 27 cases out of the 100 where the count from the 1 : 1000 

 dilution fails to check within these limits. It is important to 

 note also that 23 of these 27 cases are instances where the dis- 

 crepancy was such as to give a higher count that the accepted 

 count, indicating that chance contaminations were probably 

 the chief cause of trouble. 



SUMMARY 



1. The work here reported includes a study of the counts 

 made from 1435 agar plates inoculated from samples of market 

 milk and incubated five days at 21°C.; and also a study of the 

 counts made from 1056 of the same plates after two days addi- 

 tional incubation at 37°C. The results obtained indicate that, 

 for milk analyses, the counts made from plates having more 

 than 30 and less than 400 colonies on the plates are very nearly 

 as satisfactory as those obtained from plates having more than 

 40 and less than 200 colonies, the latter being the limits in 

 numbers originally recommended by the Committee on Standard 

 Methods for the Bacterial Examination of Milk. 



2. Plates having less than 20 or more than 400 colonies on 

 them are shown to be so frequently discrepant that counts obtained 

 from them should never be trusted unless checked by compari- 

 son with plates from different dilutions having more than 30 

 or less than 400 colonies. The acceptance of counts from plates 

 having 20 to 30 colonies per plate would not greatly increase 

 the percentage of discrepancies. 



3. All groups of plates, regardless of the number of colonies 

 showed a certain percentage of plates which gave counts which 

 varied more than 20 per cent from the accepted count. The 

 percentage of discrepant counts of this sort varied between 37 



