In recent years the paper industry has been studying different technologies to reduce fresh water consumption in the mills. In this study micro-, ultra- and nanofiltration of some paper mill waters both on a laboratory and a-pilot scale were compared. The performances of different membranes and modules are reported and some general conclusions are given. The performance of the membranes was evaluated by determining the reduction of certain measurable parameters in the feed solution such as chemical oxygen demand (GOD), total organic carbon (TOC), total solids (TS), lignin, sugar, anionic trash (CD), ions, conductivity, turbidity, silt density index (SDI) and brightness decreasing substances. The tested tubular 0.4 mu m microfiltration membrane was not suitable as a pretreatment step to nanofiltration. In ultrafiltration the tested cross rotational (CR) filter using flat sheet membranes gave significantly higher fluxes than the tested polymeric or ceramic tubular membranes. It was shown that the relatively low cut-off (30 kD) hydrophilic C30G membrane made from regenerated cellulose had higher fluxes both at neutral and acidic pH. Also the long time behaviour (duration four or six days) of the C30G was very good. It was also shown that the pH of the original feed, the experimental set-up and the membrane cut-off affected fouling the most in ultrafiltration. In nanofiltration the membrane structure and feed pH affected the results the most. Very good results were obtained by all the tested nanofiltration membranes. It was shown that fluxes similar to microfiltration are obtained by tubular ultrafiltration membranes and that the reductions were better. Moreover, the ultrafiltration process was more reliable than microfiltration as a pretreatment for nanofiltration because the obtained silt density index was always below the value 5 with ultrafiltration. However, it was also shown that in nanofiltration better reductions were obtained if no pretreatment was done, if it was not required by the module configuration.