Abstract: Equine training methods, and consequently, performance times have improved little since the last decades. With advances in measuring signals on-line by means of several new technologies and analytical procedures, and processing these signals immediately with strong and compact processors, it may be possible to develop new training methods. In this research, the objective was to explore the possibilities of using modern model-based algorithms to control the heart rate of horses (bpm) on-line by means of the control input running speed (km/h). Forty-five experiments with five horses and four riders were carried out to generate measurements of physiological status during running. The dynamical characteristics of each horse were quantified using linear discrete transfer function models. The dynamic response of heart rate to step changes in running speed were accurately described. In 90% of the cases, a first-order model gave the best fit. For 69% of the models, the r2 was higher than 0.90 and for 34% of the models, the r2 was even higher than 0.95. In a next step, the model-based algorithm was evaluated by controlling cardiac responses of two horses (horses 2 and 4) to a pre-defined trajectory. The model parameters were kept constant. On average, the error between the defined target trajectory in heart rate and the actual controlled heart rate ranged between 0.2 and 1.4 bpm for the whole target heart rate trajectory. During the steady-state part of the trajectory the average error was maximum 1.1 bpm. In the transient from one steady-state heart rate to another level, the error could increase on average up to 5 bpm. In the future, the combination of on-line measured bioresponses with real-time analysis can be used for adjusting the work load of the horse, during training, directly to the immediate needs of horse (welfare) and trainer (performance).