A hybrid coagulation-ultrafiltration process has been investigated to understand membrane performance. Coagulation prior to ultrafiltration is suspected to reduce fouling by decreasing cake resistance, limiting pore blockage and increasing backwash efficiency. Coagulation followed by tangential ultrafiltration should gather the beneficial effects of particle growth and cross-flow velocity. Our study aims at determining the key parameters to improve membrane performance, by describing floc behaviour during the hollow fibre ultrafiltration process. Flocs encounter a wide range of shear stresses that are reproduced through the utilization of different coagulation reactors. Performing a jar-test enables the formation of flocs under soft conditions, whereas Taylor-Couette reactors can create the same shear stresses occurring in the hollow fibres or in the pump. Synthetic raw water was made by adding bentonite into tap water. Five organic coagulants (cationic polyelectrolytes) and ferric chloride were selected. Floc growth was thoroughly monitored in the different reactors by laser granulometry. Coagulation-ultrafiltration experiments revealed different process performance. The effect on the permeate flux depended on the coagulant used: some coagulants have no influence on permeate flux, another enables a 20% increase in permeate flux whereas another coagulant leads to a decrease of 50%. Flocs formed with ferric chloride do not resist shear stress and consequently have no influence on permeate flux. These results show the necessity to create large flocs, but the size is not sufficient to explain membrane performance. Even if flocs show a good resistance to shear stress, a high compactness (D-f = 3) will lead to a dramatic decrease of permeate flux by increasing the mass transfer resistance of the cake. On the contrary, flocs less resistant to shear stress, then smaller and also more open have no effect on permeate flux. An optimum was quantified for large flocs, resistant enough to shear stress facilitating flow between aggregates. (C) 2008 Elsevier B.V. All rights reserved.