This research has been done using recently introduced multivariate multiscale entropy method with a view to estimating vigilance of drivers during driving in simulated environment. In this driving simulation experiment, twenty one subjects including twelve men and nine women were participated. Multivariate multiscale entropy (MMSE) has been applied to this multimodal Seed dataset for estimating vigilance from electroencephalogram (EEG) and electrooculogram (EOG) signals in order to build a vigilance detection system. The experimental MMSE analysis curves show statistically signi cant di erences (p < 0.01) in terms of complexity among brain EEG signals, forehead EEG signals and EOG signals. Moreover, the di erence in the multivariate sample entropy across all scales in awake (1.0828 0.4664), tired (0.7841 0.3183) and drowsy (0.2938 0.1664) states are statistically signi cant (p <0.01). Also, the support vector machine (SVM), a machine learning technique, has discriminated the cognitive states (awake, tired and drowsy) with the promising classi cation accuracy of 76.2%. As a result, the MMSE analysis of cognitive states can be implemented practically for vigilance detection by building a programmable vigilance detection system .