At-grade rail crossings provide different levels of warnings and/or barriers to alert drivers to the potential dangers presented by approaching trains. For some drivers, an activated warning system, rather than being a signal to stop, merely serves as a cue for the need to make a decision whether or not to cross. In California, for the ten-year period from 2001 to 2010, the result was 1,033 trainvehicle crashes resulting in 157 deaths and 458 injuries. The best solution to rail crossing crashes is to remove the need for the driver to engage in a potentially faulty decision-making process by making it impossible, or at least very difficult, for the driver to bypass lowered gates. Two low-technology, low-cost, low-maintenance methods: median separators and long-arm gates, have been deployed in many locations and have been shown to prevent deaths and injuries while remaining economically feasible. Highway-railway grade crossing collisions tend to be spread over a vast number of sites, with few (if any) occurring at any given site in any given year. To improve safety at all 6,443 grade crossings in California to some uniform standard would be prohibitively expensive and impractical. Therefore, any comprehensive safety program must begin by first identifying crossings where the risk of collision is unacceptably high, and where safety countermeasures are most warranted. Predicting the degree of safety present at highway-railroad grade crossings using accident prediction models is a common approach. These models are usually developed using a highwayrailroad grade crossing database (primarily that maintained by the Federal Railroad Administration using data supplied by each state) consisting of crossing characteristics and accident data for a given period of time. From the perspective of the California Division of Rail, however, the search for the ideal formula or ranking system is immaterial given the current state eScholarship provides open access, scholarly publishing services to the University of California and delivers a dynamic research platform to scholars worldwide. of its rail crossing inventory database (which is the responsibility of the California Public Utilities Commission) with its often inaccurate as well as incomplete information. At present, the only meaningful statistic for which there is data is crash history, which leads to the difficult question of what constitutes a dangerous rail-highway crossing. If a crossing has an incident every twenty years, it would be hard to argue that the crossing is dangerous. But what if a crash occurs there every ten years, or every five years? At what point does a crossing become dangerous and in need of remedial action? An examination of gated sites with multiple crashes between 1986 and 2010 shows that the median time between crashes ranges from 6.1 years at sites with two crashes to 2.75 years at sites with five crashes. Also of interest is the length of time it took each site to accumulate its crashes. These range from 5 days to 24 years for sites with two crashes to 8 to 23 years at sites with five crashes. Certainly a crossing with five incidents over the course of eight years can be labeled as dangerous, but what about those crossings with incidents spread out over twenty or more years? If a crossing is truly dangerous and in need of remedial action, how do so many vehicles make it safely across resulting in several years passing without an incident? Since so much depends on the accuracy of our state’s inventory database, bringing it up to date and putting it into a readily accessible format should be the top priority for all involved in California rail. Once that is accomplished, crossings will be able to be properly evaluated by looking for commonalities at sites where crashes have occurred in the past.