Many drunk drivers are probably escaping detection and prosecution in fatal accidents because of procedural differences among police departments and the relative importance they give to traffic deaths and drunk driving. Studies suggest that drivers are most likely to be tested if they show visible signs of intoxication, even though drinkers in the middle stages of alcoholism may function very well at high blood alcohol content levels. In addition, the shock reaction to an accident can pump adrenalin into the driver, thus altering alcohol's apparent effects. Until 1985, estimates of drunk driving deaths nationwide were based upon tests of decreased drivers in only 15 States. The new method is an improvement, but still assumes that alcohol involvement is equally likely in tested and untested drivers. Some States perform almost no testing, and a 1983 study showed that 40 States failed to comply with their own laws requiring such tests. The report recommends universal testing of all drivers in fatal accidents. Five references, tables, and the Minnesota study are provided.