A new machine vision approach for estimating, monitoring, and controlling manufactured product appearance is illustrated. This new approach consists of the following: (1) extraction of textural information from product images, (2) estimation of measures of the visual quality of the product from the textural information extracted, (3) modeling causal relationships between the estimated quality and process variables, and (4) optimization of new operating conditions using the causal model. This method is specifically aimed at treating the stochastic nature in the visual appearance of many manufactured products. This nondeterministic nature of product appearance has been a main obstacle for the success of machine vision in the process industries. This approach is successfully applied to an industrial process for estimation, modeling and optimization of the visual appearance of injection-molded plastic panels.