Abstract The neural network technique was applied to the study of the crossflow microfiltration process. Two application procedures are presented: (i) the “black-box” approach does not require an accurate description of the process, relying merely on the ability of neural networks to approximate the dynamics of any system; (ii) the semi-physical approach is an attempt to take into account a priori knowledge. Neural networks are then used simply to assess unknown parameters. Experiments were performed on suspensions of baker's yeast. In order to obtain the data set necessary required to train the different networks, two concentrations were tested in several operating conditions (filtration pressures and tangential flow velocities).