Mooney viscosity is an essential parameter in quality control during the production of nitrile-butadiene rubber (NBR) by emulsion polymerization. A process model that could help understand the influence of feed compositions on the Mooney viscosity of NBR products is of vital importance for its intelligent manufacture. In this work, a process model comprised of a mechanistic model based on emulsion polymerization kinetics and a data-driven model derived from genetic programming (GP) for Mooney viscosity is developed to correlate the feed compositions (including impurities) and process conditions to Mooney viscosity of NBR products. The feed compositions are inputs of the mechanistic model to generate the number-, weight-averaged molecular weights (M_n, M_w) and branching degree (BRD) of NBR polymers. With these generated data, the GP model is used to output the optimal correlation for the Mooney viscosity of NBR. In a pilot NBR production, Mooney viscosity data of NBR predicted by the process model agree quite well with experimental values. Furthermore, the process model enables the analyses of the univariate and multivariate influence of feed compositions on NBR Mooney viscosity, and the variables include the contents of vinyl acetylene and dimer in 1,3-butadiene, as well as the mass flow rate of the chain transfer agent (CTA) in the process. Based on the results, it is recommended to control the content of vinyl acetylene in the 1,3-butadiene feed below 14 ppm and the content of dimer below 1100 ppm. This developed process model would help stabilize NBR viscosity for a better control of the product quality.