A major cause of vehicle accidents is driver impairment due to drowsiness. Further research is needed on drowsiness definition and its countermeasures. Four studies were recently undertaken, two of which have dealt with problems of drowsiness definition while two have dealt with on-line drowsiness detection. The first definitional study involved observer rating and the second involved prediction of reductions in task performance based on combinations of physiological measures. Both approaches appear promising. These definitions, as well as others, could then be related to measures taken from the vehicle itself (e.g., measures of steering, lane position, and lateral acceleration), which could be combined to form drowsiness detection algorithms. The first detection study involved algorithm development and expected accuracy, and the second involved validation of typical algorithms. This paper provides an overview of the four studies and an assessment of the feasibility of on-line detection of drowsiness.