Reverse osmosis (RO) has found extensive usage in the fields of desalination and pollution control. In the present work;,an attempt is,made to model the separation of sodium chloride-water system by reverse osmosis using neural nets. Experimental data are used to train the network developed for the said system. The training data included the feed concentration range from 1000 to 30000 ppm, pressure range from 20 to 100 atm, and feed rates from 300 to,1500 mL/min. The network thus, developed has been found to predict the system variables within the error range of +/-1%, except for sudden deviations in:process parameters, and initial and final, value of flow rates at low pressures. Nearly the same trend, that is, maximum errors at lower pressures and higher flow rates, were observed in the prediction of RO performance with membrane transport models, reported earlier. The reasons for this may be the-unsteady state behaviour of the system, or system instability or error in experimentation. Such deviations are hot of much importance, because the predicted and experimental values are within the satisfactory range.