Amyloglucosidase (AG) and invertase (INV) were immobilized into polyacrylonitrile (PAN) and carboxyl-modified polyacrylonitrile (PAN-AA) ultrafiltration membranes. The effects of membrane pore size and immobilization method on total enzymic activity, enzyme distribution across the membrane structure, and long-term stability were investigated. Immobilization by adsorption did not yield stable enzyme membranes. Enzyme crosslinking with glutaraldehyde and covalent binding after carboxyl activation with water-sol. carbodiimide were proven to be suitable methods to yield enzyme membranes with stable activities (about 1 U cm-2 AG bound to the high sp. surface area, low permeability membranes and 0.055 U cm-2 AG and 2.3 U cm-2 INV bound to the low surface, high flux membranes). Covalent immobilization yielded AG UF membranes with no activity drop within one year. The hydrophilic surface of PAN-AA membranes reduced unspecific enzyme adsorption. The immobilization reaction was driven by the reactivity of the activated surface. The accessibility of the membranes and the substrate size (starch vs. maltose for AG) had a major impact on obsd. activities. UF membranes with apparently \"sym.\" enzyme distribution were obsd. for moderate total loadings if sufficient substrate access on both outside surfaces is provided. The Michaelis-Menten const. Km with the small substrate maltose were 3.5 mmol-1.l for free and 3.2 mmol-1.l for covalently bound AG whereas Vmax values with maltose and starch, dropped to 32% and 22%, resp., for the immobilized vs. free enzyme. Under ultrafiltration conditions with a transmembrane pressure of 0.3 MPa, the INV membrane activity rose to 250% compared with tests where substrate transport occurred by diffusion exclusively.