The stress-bearing fabric of bacteria is made of peptidoglycan. This crosslinked fabric is formed from disaccharide pentapeptide units that are transported through the cytoplasmic membrane and then polymerized in two directions: (i) to form oligoglycan chains; and (ii) to cross link these chains by tail-to-tail bonds from the muropeptides to the protruding peptides of other chains. The distribution of the glycan chain lengths is reminiscent of the &&most probable distribution'' of polymer chemistry. Of course, the process is more complex than solely the random addition of units to growing chains. The complexity precludes mathematical analysis, but computer modeling of the Monte Carlo type is capable of including a range of possibilities. At each time point a speci"ed number of disaccharides are singly added to the muramic acid residue ends of existing chains chosen at random. The transfer is in exchange for the cleavage of pyrophosphate bactoprenol that transported the disaccharide pentapeptide through the membrane. The progam then selects, again at random, which chain to cleave and between which two disaccharides of the chain the cleavage event is to occur. The cleavage generates an N-acetyl 1,6 anhydro-muramic acid end and a non-reducing N-acetyl glucosamine end. The simulation can be modi"ed so that the program does not cleave o! a disaccharide next to either end of the chain. Comparisons are shown with the experimental results of Obermann & HoK ltje (1994. Microbiology 140, 79}87.) They obtained their data by taking the results with normal growing cells and subtracting the similar data from minicells to estimate the chain length distribution in the cylinder part of the cell. In its most basic form the computer simulation has only one "tted parameter, K, which is the number of disaccharides added to the murein for every internal cleavage event. In this form the "tting to the experimental results is poor. One possible reason for this is that the tension on the chains, and therefore the probability of being cleaved by autolysins varies with orientation of the chain on the cylinder surface. It is well known that the tension in the cylindrical wall is twice as large in the circumferential direction as in the axial one, so one class would consist of those chains aligned longitudinally, subject to lower stress, and would have a higher energy of activation for autolysis than chains aligned circumferentially. A good "t is obtained on the assumption that there are only two classes of chains; one more likely to be cleaved than the other. The key point is that only two processes: adding of disaccharide pentapeptides at random to glycan chains and cleavage between the disaccharides at random, together with the assumption that the wall is less easily hydrolysed in the axial direction is su$cient to account for the experimental distribution.