During membrane-based separation of proteins, changes in protein concentration of the permeate and retentate streams occurs over time. The current work proposes a new approach for monitoring the changes in concentrations of proteins in both permeate and retentate by making use of data collected using fluorescence spectroscopy and intrinsic protein fluorescence analyzed by multivariate statistical techniques. Whey protein isolate consists mainly of alpha-lactalbumin (alpha-LA), beta-lactoglobulin (beta-LG), and small proportion of bovine serum albumin (BSA) and was used as a model system in this study. A fiber optic probe (FOP) was used to acquire multiwavelength fluorescence spectra for permeate and retentate streams at different times during UF-based separation of the components from a multicomponent solution. Multivariate regression models were developed for predicting the concentrations of alpha-LA, beta-LG, and BSA by establishing a calibration model between data acquired using the FOP and the corresponding protein concentration levels measured by size-exclusion chromatography. The model was validated using FOP data that were not previously used for calibration of the regression models. This comparison showed that concentrations of alpha-LA, beta-LG, and BSA could be predicted directly from FOP data within reasonable accuracy by making use of multivariate calibration tools. This approach has several attractive features including that it is nondestructive, fast, and relatively simple to perform. This technique has potential practical applications as it could offer the opportunity for in situ monitoring of membrane filtration processes by tracking individual protein transmission and selectivity of fractionation. (C) 2009 American Institute of Chemical Engineers Biotechnol. Prog., 26: 168-178, 2010