Purified rat-liver Golgi stacks were extracted in a buffer containing 2% (w/v) TX-100, 50mM MOPS pH7.0, 0.1mM MgCl2, 1mM DTT and 10% (w/v) sucrose, and centrifuged to produce an insoluble pellet which contained the majority of three medial-Golgi enzymes, mannosidase II and N-acetylglucosaminyltransferases I and II. Proteins from other regions of the Golgi stack were mostly solubilised. A further extraction of this pellet in the TX-100 buffer containing 150mM NaCl led to complete solubilisation of these medial-Golgi enzymes. After the salt extraction, a second insoluble pellet was produced which was termed the Golgi matrix. The salt-solubilised medial-enzymes could rebind the matrix upon dialysis, while an enzyme from the trans-Golgi could not. Scatchard analysis revealed that rebinding was saturable and occurred with a high affinity, suggesting that the matrix contained a fixed number of receptors which specifically bound the medial-enzymes. This suggested that the enzyme insolubility in detergent was due to their interaction with the matrix. Digestion of intact Golgi membranes with proteinase K greatly increased the detergent-solubility to the medial-enzymes, suggesting that components of the matrix were present on the cytoplasmic, intercistemal face of the Golgi membranes. This topological orientation suggested that the matrix might play a role in stacking the Golgi cisternae. Binding of the enzymes did not, however, occur via their cytoplasmic or membrane-spanning domains, suggesting that the matrix is a complex structure, containing components on both sides of the Golgi membrane. Because of its topology and its binding capacity for medial-enzymes, the matrix may function in aiding the retention of Golgi proteins, maintaining the flattened cisternal morphology or in connecting adjacent cisternae to form the characteristic Golgi stack.