We present a model of cognitive workload based on the ACT-R (Adaptive Character of Thought - Rational) cognitive architecture. That model has been validated in a synthetic air traffic control task according to a wide range of behavioral measures. Its cognitive workload predictions are sensitive to level of task embedding, interaction speed, level of interface decision support and individual differences. We sketch possible extensions of the model that support multiple workload dimensions and instantaneous workload ratings. We also discuss possible applications of this kind of fine-grained computational models to system design, operator training and selection and dynamic load balancing.