This paper describes the development of inferential models for the provision of real-time, on-line estimates of the quality of a breakfast cereal for production line operators. Five quality variables were selected and on-line measurements reflective of the key process conditions were identified. Following process data logging, a number of linear and non-linear data-based modelling methods were applied to identify relationships between the on-line measurements and the product quality. Off-line verification of the models indicated that the prediction accuracy achieved was sufficient to offer the opportunity for quality control improvements. The models were subsequently implemented on-line to provide the process operators with frequent estimates of product quality. Performance assessment has indicated a reduction in the variability of all five quality parameters. In addition to details of the modelling, the decisions relating to the development strategy and justification for implementation are considered.