Four sets, each of 100 bacterial isolates, were derived from two different flax rhizosphere and corresponding control (nonrhizosphere soil) samples. A two-stage principal component analysis was used to compare the nature and properties of these bacterial cultures which were selected by a random sampling method from primary isolation plates. These 400 experimental isolates were characterized by their reaction to a set of 98 tests. Principal component analysis was applied separately to the data for each of the four sets of cultures. The sets were then compared both in terms of the leading discriminating tests and of the detailed properties of the clusters of isolates detected. Because these isolates and a set of 100 named cultures (previously studied) were subjected to the same tests, the properties of groups of isolates from different sets could be directly compared. The kinds of isolates predominant in the rhizosphere samples were not represented amongst isolates derived from the corresponding control soil populations. The specific kinds of amino-acid-requiring isolates present in the control soil populations were not found in the corresponding rhizosphere populations. The two control soil sets of isolates (predominantly arthrobacter–coryneform-like organisms) were generally similar in their reactions to the tests used, but the two sets of rhizosphere isolates (predominantly pseudomonas-like organisms) were relatively dissimilar. The named species studied previously were not sufficiently indicative of the diversity found in the experimental isolates: this was especially true of the arthrobacter-like isolates. The numerical analyses yielded useful groupings of tests and of cultures and aided in their evaluation.