This paper proposes a binaural model for estimating the sound image width of the audio-listening environment in a car cabin. Based on a binaural model for the sound image azimuth estimation proposed in a previous study, a new feature was extracted and expanded into a model capable of estimating the sound image width. To improve the estimation accuracy of this model, the width of the sound image azimuth estimation distribution was extracted from the attack region of the gammachirp auditory filter bank outputs, and a multiple nonlinear regression model that uses the feature and sound image azimuth estimate value was constructed and examined. As the results indicate, the proposed method achieves a significantly higher accuracy.