Two parallel membrane bioreactors (2m³ each) were operated over a period of 2 years. Both pilots were optimised for nitrification, denitrification, and enhanced biological phosphorous elimination, treating identical municipal waste water under comparable operating conditions. The only constructional difference between the pilots was the position of the denitrification zone (pre-denitrification in pilot 1 and post-denitrification in pilot 2). Despite identical modules and conditions, the two MBRs showed different permeabilities and fouling rates. The differences were not related to the denitrification scheme. In order to find an explanation for the different membrane performances, a one-year investigation was initiated and the membrane performance as well as the operating regime and characteristics of the activated sludge were closely studied. MLSS concentrations, solid retention time, loading rates, and filtration flux were found not to be responsible for the different performance of the submerged modules. These parameters were kept identical in the two pilot plants. Instead, the non-settable fraction of the sludges (soluble and colloidal material, i.e. polysaccharides, proteins and organic colloids) was found to impact fouling and to cause the difference in membrane performance between the two MBR. This fraction was analysed by spectrophotometric and size exclusion chromatography (SEC) methods. In a second step, the origin of these substances was investigated. The results point to microbiologically produced substances such as extracellular polymeric substances (EPS) or soluble microbial product.