The voltage-dependent kinetics of sodium channels are a critical factor in the generation of action potentials in pacemaker neurons, and in modulating their firing properties. However, it is not only difficult to build detailed models of Na channel gating, but it is even more difficult to predict neuronal behavior, when the other ionic currents are generally unknown. A solution to this problem is the dynamic clamp technique, which can be used to replace a voltage-gated ion channel with a computational model. Effectively, the neuron and the computer become a hybrid biological-computational simulator, allowing testing an ion channel kinetic model without any knowledge about other currents. Moreover, the model can be optimized in real-time, by varying its kinetic properties in order to match the neuronal behavior. In this talk, I will present how we used the dynamic clamp technique to understand the role of voltage-gated sodium channels in two types of slow pacemaker neurons: medullary serotonergic raphé neurons, and respiratory neurons in the pre-Bötzinger complex. First, we constructed Markov models from voltage clamp recordings, based on the topology proposed by Kuo and Bean, with the addition of states to account for slow properties (inactivation and recovery from inactivation). Then, the models were tested with dynamic clamp in neurons, with a focus on the interaction between subthreshold current and slow inactivation, and spiking properties. The key features of our Windows-based dynamic clamp system will also be reviewed, including the procedure for offline and online fitting. The program exploits the multiprocessor architecture to obtain excellent real-time, multi-tasking performance. For example, a Markov model with 14 states can be solved at almost 100 kHz. At the same time, data are visualized in real-time and continuously saved to disk, and additional computation can be run in parallel.