Size does matter — at least as far as groups are concerned. In very small groups, the addition or loss of one member can of course make a radical difference to the group process. Larger groups need to be managed in quite different ways from smaller ones. So let's tackle this systematically: "Formal" features refer to necessary properties of the group, and are functions of the number of people: "Process" features are more empirically determined, and assume more importance as the size gets greater.
But what is a "group"? For present purposes, let us stipulate that it is:
 a collection of people
 in each others' presence (I am leaving aside virtual groups and sociologically defined reference groups here: this site is mainly about facetoface teaching, after all)
 who are aware of each other (some of my students make the sound point that a lot of the theory breaks down in the case of people with severe or profound learning disabilities. Note, too, that beyond "awareness", the manner of awareness can also make an important difference, as in the case of sensory impairments: eyecontact is not available as a "cueing" device, for example, for visuallyimpaired people—and the process of communication can be radically changed if the medium is visual, as with sign language)
 and who interact with each other (the collection of people on a bus, or in a doctor's waiting room, may simply be regarded as a collection of individuals — or a set of small, dyadic or triadic, groups — for most purposes and under normal circumstances)
1  Can you have a group of one? Not formally, but in process terms the question raises a number of issues which are worth thinking about:
All these scenarios suggests that the group has emergent properties which go beyond the individuals who comprise it. 
2  Formally: potential for unanimity or equal disagreement. Each person potentially has 50% of the "airtime". Very vulnerable to loss and addition.
Process: the mating group, with potential for very strong feelings. The irreducible minimum for emergence of any group properties. Power becomes an issue. Polarisation and projection are possible. 
3  Formally: potential for unanimity or majorityminority splits or fragmentation. Each person potentially has 33% of the "airtime". Very vulnerable to loss and addition.
Process: now the group is getting more interesting. 
4  Formally: potential for unanimity, equal splits or threeversusone splits. Each member could have 25% of airtime. Slightly less vulnerable.
Process: Variety makes complete fragmentation less likely. This is probably the optimum size for small syndicate or "buzz" groups: there is sufficient variety in the group to reap the benefits of group working, but it is not large enough for anyone to hide. Subgroupings of any substantial duration are possible, but not very likely. 
56 
Formally: less change this time, except that with five members you can only have equal splits if someone is not counted or abstains. The possibility of four or fiveversusone splits makes group pressure more potent. Airtime now down to 20%, which is low, considering the process issues of participation levels.
Process: Can be used for exercises as above, but roles may be more discrete and are slightly more likely to get "fixed". Either bring in a formal — if lightweight — structure, such as suggesting a chairperson, or use when the group is to act as a team with predetermined roles. 
78 
The preferred size for therapy groups.

12 
The next qualitative shift comes with the upper limit of the "small group". Beyond twelve, the avoidance of subgroupings is only possible with quite active structuring.
This is a reasonable size for a participative class, however. It is divisible by 2, 3, 4 and 6, offering useful combinations, and the role of the teacher as chairperson is sufficient to address its normal problems. 
18 
Allowing for resource constraints, the next best size for a class. All the potential problems of 12 are amplified, but still quite manageable, and you have almost the same factoring possibilities for small group working, although you will have to settle for two fours and two fives for syndicates: even so, there aren't too many groups to take reportingback.
However, there is a strange phenomenon I have noticed over many years: it is almost impossible to set up a short course with exactly 18 members attending. Either someone drops out if you recruit exactly the right number, or if you recruit over to be on the safe side, they all turn up. 17 and 19 are both prime numbers, and really foul up any pairbased work! 
30

Let's get real. Class sizes in compulsory education are at this level or greater in the state sector, and in higher education it is not unknown to have socalled "seminar" or even "tutorial" groups of this size. That calls for another page. 
This is an archived copy of Atherton J S (2013) Learning and Teaching [Online: UK] Original material by James Atherton: last updated overall 10 February 2013
This work is licensed under a Creative Commons AttributionNoncommercialNo Derivative Works 3.0 Unported License.
"This site is independent and selffunded, although the contribution of the Higher Education Academy to its development via the award of a National Teaching Fellowship, in 2004 has been greatly appreciated."