Membrane-bound enzymes from developing legume-seed endosperms catalyse galactomannan biosynthesis in vitro from GDP-mannose and UDPgalactose. A mannosyltransferase [mannan synthase] catalyses the extension of the linear (l~4)-~-linked Dmannan backbone towards the non-reducing end. A specific ~-galactosyltransferase brings about the galactosyl-substitution of the backbone by catalysing the transfer of a (1 ~6)-~-D-galactosyl residue to an acceptor mannosyl residue at or close to the non-reducing terminus of the growing backbone. Labelled galactomannans with a range of mannose/galactose (Man/Gal) ratios were formed in vitro from GDP-[14C]mannose and UDP-[14C]galactose using membrane-bound enzyme preparations from fenugreek (Trigonella foenum-graecum L.), guar (Cyamopsis tetragonoloba (L.) Taub.) and senna (Senna occidentalis (L.) Link.), species which in vivo, form galactomannans with Man/Gal ratios of 1.1, 1.6 and 3.3 respectively. The labelled galactomannans were fragmented using a structure-sensitive endo-(1 ~4)-]3-D-mannanase and the quantitative fragmentation data were processed using a computer algorithm which simulated the above model for galactomannan biosynthesis on the basis of a second-order Markov chain process, and also the subsequent action of the endo-mannanase. For each galactomannan data-set processed, the algorithm generated a set of four conditional probabilities required by the Markov model. The need for a second-order Markov chain description indicated that the galactomannan subsite recognition sequence of the galactosyltransferase must encompass at least three backbone mannose residues, i.e. the site of substitution and the two preceding ones towards the reducing end of the growing galactomannan chain. Data-sets from the three plant species generated three distinctly different sets of probabilities, and hence galactose-substitution rules. For each species, the maximum degree of galactose-substitution consistent with these rules was closely similar to that observed for