In this work a non-steady state model for cross flow ultrafiltration was tested and the results estimated by the model were compared with those experimentally obtained. Crossflow ultrafiltration experiments were performed in a pilot plant using tubular ZrO2-TiO2 ceramic membranes with a MWCO of 15 kDa (Orelis, France). The membranes were fouled with a 5 gAL PEG (35000 Da molecular weight) aqueous solution at different transmembrane pressures and crossflow velocities. The main goal of this work was to predict flux decline with a model whose input parameters are dependent on the operating conditions. Therefore all the model parameters were theoretically calculated, avoiding the utilization of empirical parameters. The validity of the model tested was observed to be restricted to the case of low crossflow velocities where gel layer formation is the main fouling mechanism.