Reactive extrusion processes are typically multivariable, display highly nonlinear characteristics, and often have significant time delays associated with the (offline) measurements of key product properties. Achieving desired product characteristics in industrial practice has therefore been based primarily on the control of a single critical variable such as viscosity. However, increasingly stringent customer demand on product quality has rendered such strategies no longer viable and has necessitated the development of more comprehensive schemes that focus explicitly on controlling product quality characteristics. This paper reports on an experimentally validated inference-based control scheme for controlling product quality and end-use properties in reactive extrusion processes. The scheme employs inference models to predict infrequently measured properties at a much faster rate—predictions that are then used to take necessary control action in between samples, within a cascadelike structure involving separate and distinct multivariable controllers. The control scheme is evaluated first in simulation and then implemented experimentally via a Labview−Matlab interface on an actual pilot-scale reactive extrusion process, where product viscosity, tensile strength, and toughness are controlled simultaneously. Some representative results are presented to highlight the advantages and limitations of the scheme.