Variability of Bacteria in North Atlantic Sediments 523 



ing a sterile glass tube vertically into the surface of the core. Tlie 

 bore of the tube was known, hence tlie volume of the sample was 

 determined from its length. The samples were either immediate!) 

 prepared for counting or frozen in the glass tubes for later study. 

 Dilutions of 250 or 2500 fold were made in 80 per cent sea 

 water; the lower dilution tends to set a limit for the method be- 

 cause of the amount of sediment deposited upon the filter. The 

 medium recommended by Carlucci & Framer ( 1 ) was used in 

 liquid form, at double concentration, and supplemented by vita- 

 min B12. Further details of the counting technique are given in 

 Hayes & Anthony (4). Direct counts were made by fluorescence 

 microscopy (6). An ordinary household "deepfreeze" was used 

 to freeze and store sediment samples. 



RESULTS 



Precision of Counting Bacteria in Marine 

 Sediments by Membrane Filters 



Under ideal conditions, plate counts of bacteria constitute a 

 Poisson series (3). It is characteristic of the Poisson distril)ution 

 that the variance (standard deviation squared) is equal to the 

 mean. Advantage may be taken of this characteristic to check 

 the goodness of fit between observed and theoretical numbers. 

 For lake sediments, Hayes & Anthony (4) plotted the standard 

 percentage error of the mean against the number of colonies per 

 filter and compared the observed values with the Poisson expecta- 

 tion. This analysis has been repeated for 117 quadruplicate filter 

 counts of marine bacteria. The results are shown in Figure lA. 



The Poisson series describes a situation in which it is pre- 

 sumed that all error is inherent, i.e., due to the random distrilni- 

 tion of objects in time or space. Counts of biological material ma) 

 be expected to incur additional error due to handling. In Figure 

 IB, the line describing the inherent error of the Poisson series is 

 dependent upon the mean, i.e., it forms a rectangular hyperbola 

 falling from high values associated with low mean counts toward 

 infinitely small values as tlie mean increases (within the practical 

 range shown here, the error tends to le\'el off at about ± 3%). 

 Handling error, on the other hand, should be more or less con- 

 stant and independent of the mean count. Figure IC is a plot of 



