A new model-based predictive technique is proposed for generating early warnings to operators. The proposed method uses an open-loop process and disturbance model along with process measurements to predict the future state of a process. The method successfully monitors a process vulnerability (i) during the time-delay period and (ii) when the actuator does not have sufficient capacity. On the basis of the predicted process system, an alarm is generated when predictions exceed the threshold of safety limits. A moving horizon predictor combined with bias corrections predicts the dynamic output for the time-delay period, and a linear programming algorithm checks the feasibility of the solution using system gain and constraint information. The proposed method is applied to a single-input-single-output system as well as a multiple-input-multiple- output continuous stirred tank heater system. The method demonstrated performance in issuing early warnings that was superior to that of existing methods and was robust under small disturbances in the process.