1. Background Most of the driving assist systems are uniformly controlled without considering the difference of the individual driver’s characteristics. Drivers feel discomfort, nuisance and stress, because the function of system is different from the driver’s characteristics. This research aims to reduce these side effect for the systems with a high accuracy driver model. This model is constructed by NARX(Nonlinear AutoregRessive with eXogeous inputs) that has a learning function and can estimate the driving action by a driver. A new point of this research investigated that the model was constructed by one driving condition, and this model can be applied to the other driving conditions. If one of the model can apply to many driving conditions, a system can construct as minimum requirements. 2. Experimental method The driver decelerated with the target at the end of the traffic jam in highway. A driver model constructed for the driver’s braking action. The experimental conditions were eleven data measurements from 50km/h to 130km/h every additional 10km/h. A model was constructed with datas which is from one data to ten datas. From this analysis, the number of data clarify for constructing the model. And the accuracy of the model was confirmed from 50km/h to 130km/h every 10km/h, this analysis investigated that the model of accuracy increase when the data has the difference of velocity. 3. Conclusion The accuracy of the model improved when the number of learning data was increased. It is clarified the relationship between the number of data and the model’s accuracy. In case of the difference of velocity, the accuracy decreased when the difference increased, and this tendency is more shown when the host vehicle drove at a low velocity. However this tendency was less affected at high velocity.