Drunk driving remains a major factor in fatal accidents around the world. Previous studies have suggested various approaches to prevent drunk driving. For example, sensors were developed to detect alcohol intoxication. Other significant studies has shown how alcohol intoxication affects driving maneuvers and the resulting vehicle behavior. Such approaches have high applicability because they do not require the installation special devices. In addition, alcohol is known to induce drowsiness. Drowsiness also affects driving maneuvers and vehicle behaviors. Therefore, to accurately detect alcohol intoxication, the effect of drowsiness must be separated from that of alcohol intoxication. However, the difference between these two types of impaired driving remains unclear. If the type of impaired driving can be identified, effective countermeasures may be applied. To address this issue, the present study distinguishes between drunk driving and drowsy driving based on driving maneuvers and the resulting vehicle behavior, which are determined using data from vehicle-based sensors. Data on driving maneuvers and vehicle behavior were collected using a driving simulator set to a simple driving scenario to induce drowsiness. The experiment consisted of five driving sessions. The first session was sober driving (i.e., before drinking). Following the first session, the participants took a meal with an arbitrary amount of alcoholic drink. From the second session to the fifth sessions were drunk driving. Breath alcohol concentration (BrAC) was measured and driver drowsiness was determined by using the Stanford Sleepiness Scale (SSS). The results which culculated by standardized partial regression coefficient suggest that the standard deviation of the steering angle is affected by alcohol, whereas the standard deviation of lateral vehicle position is affected by drowsiness. Discriminant analysis were used for discriminating between four states: “Sober and Awake”, “Sober and Drowsy”, “Drunk and Awake”, and “Drunk and Drowsy”. The “Sober and Awake” state is accurately detected at a rate of 96.8%; for “Drunk and Drowsy” the rate is 65.1%, and for “Drunk and Awake” the rate is 41.1%. We discuss how, in general, almost all vehicle behavior reflects the driver’s maneuvers. For example, alcohol intoxication led to a relatively large increase in the standard deviation of steering angle, although the standard deviation of lateral vehicle position increased only slightly. Conversely, drowsiness led to a relatively small increase in the standard deviation of steering angle but a relatively large increase in the standard deviation of lateral vehicle position. This mismatched relation may be caused by frequencies involved in steering: high-frequency steering results from alcohol intoxication, whereas low-frequency or intermittent steering results from drowsiness. This means that high-frequency steering is poorly reflected in vehicle behavior because of the integral characteristics of the vehicle steering mechanism. We conclude that alcohol intoxication mainly affects driving maneuvers whereas drowsiness mainly affects vehicle behavior. Moreover, the normal state “Sober and Awake” and the impaired state “Drunk and Drowsy” may be discriminated based on these evaluation indices. Future work should investigate the frequency of steering operations in each driving state. These findings should be integrated into advanced driver assistance systems to assist impaired drivers.