Research conducted over the past three years in the Vehicle Analysis and Simulation Laboratory at Virginia Tech has resulted in the development and validation of algorithms for the detection of driver drowsiness. Specifically, the goal of the research has been to develop the best possible drowsiness-detection algorithms using measures that can be computed while a vehicle is in motion with minimal interference with the driver. The results of these studies, which have been previously reported, generally support the feasibility of drowsy-driver detection and indicate that further analysis and refinement of the algorithms is warranted. This thesis researches several methods of refining existing driver-status algorithms, the integration of driver-performance deterioration measures, and the selection of appropriate alarm thresholds to be used in test and evaluation study. The results of five algorithm optimization refinements are described. Chapter 2 reports that the elimination of outlier dependent measure data prior to algorithm development was found not to improve algorithm accuracy. Chapter 3 describes that the addition of cross product and squared terms to the algorithms did not provide consistent improvement in algorithm accuracy. Chapter 4 reports that, although time-on-task variables were found to have some improved capability, they did not consistently add to the accuracy of the algorithms. / Master of Science