BACKGROUND The bottleneck for real time control and real time release is lack of product-specific in-line sensors or fast at-line methods suitable for model-based prediction of process outcome. The most common sensors for protein purification are UV absorbance values measured at 280 and 260 nm. They have very high selectivity for proteins which contain aromatic amino acids. The 260 nm signal is more selective for nucleic acids. This work addresses the question if osmolality can be used as an additional predictor for protein purification. RESULTS An antibody intermediate purification step in flow-through mode was evaluated. The flow-through fractions were collected and then subjected to analysis for antibody concentration and osmolality. UV280, UV260, UV214, pH and conductivity have been measured on-line by the chromatography workstation. Different combinations of on-line sensor signals and osmolality have been used to find out if molality is a valuable predictor. The root mean square error was used for assessing the quality of the model-based prediction of quantity with partial least squares in this chromatography process. Predictors UV280, UV260, UV214, pH and conductivity showed equal root mean square error (0.274) as UV280, UV260, UV214, pH, conductivity and osmolality (0.274). Lowest mean square error (0.244) was found with UV280, UV260 and osmolality as predictors of quantity. CONCLUSION Osmolality as an at-line method is an excellent predictor together with UV280 and UV260 for protein quantity in model-based prediction using partial least squares methodology. (c) 2019 The Authors. Journal of Chemical Technology & Biotechnology published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.