The lengths of bacterial polysaccharides can be critical for their biological function. Unlike DNA or protein synthesis, where polymer length is implicit in the nucleic acid template, the molecular mechanisms for regulating polysaccharide length are poorly understood. Two models are commonly cited: a “molecular clock” regulates length by controlling the duration of the polymer extension process, whereas a “molecular ruler” determines length by measurement against a physical structure in the biosynthetic complex. Escherichia coli O9a is a prototype for the biosynthesis of O polysaccharides by ATP-binding cassette transporter-dependent processes. The length of the O9a polysaccharide is determined by two proteins: an extension enzyme, WbdA, and a termination enzyme, WbdD. WbdD is known to self-oligomerize and also to interact with WbdA. Changing either enzyme’s concentration can alter the polysaccharide length. We quantified the O9a polysaccharide length distribution and the enzyme concentration dependence in vivo, then made mathematical models to predict the polymer length distributions resulting from hypothetical length-regulation mechanisms. Our data show qualitative features that cannot be explained by either a molecular clock or a molecular ruler model. Therefore, we propose a “variable geometry” model, in which a postulated biosynthetic WbdA-WbdD complex assembles with variable stoichiometry dependent on relative enzyme concentration. Each stoichiometry produces polymers with a distinct, geometrically determined, modal length. This model reproduces the enzyme concentration dependence and modality of the observed polysaccharide length distributions. Our work highlights limitations of previous models and provides new insight into the mechanisms of length control in polysaccharide biosynthesis.