Fouling, a mass transfer phenomenon that depends on the solubility of the proteins, in food industry is a severe problem causing loss of efficiency of heat exchangers. In this work, ovalbumin and beta-lactoglobulin fouling was studied through of protein solubility curves. Mathematical simulation using artificial neural networks with several configurations containing independent variables (velocity, convection heat and mass transfer coefficients, temperature, potential of Hydrogen and molecular descriptors) and dependent variables (percentage of fouling removed and the reduction of free flow area) was made. Results showed that the thickness of the fouled deposit is inversely and directly proportional to the Reynolds number and the inlet temperature, respectively. For transient and turbulent flows, egg white fouls faster than milk. The fouling was greater at pH 5.0, for beta-lactoglobulin. Results obtained by the artificial neural networks were successful performance for simulating the fouling phenomenon for both natural proteins for the architecture 5-20-2, where the input layer has 5 neurons, the hidden layer has 20 neurons and the output layer has 2 neurons. (C) 2018 Elsevier Ltd. All rights reserved.