The Storm Prediction Center's tornado database is maintained and annually updated. It now contains 51 years of information on all reported United States tornadoes. During that period, there have been many changes in tornado reporting efficiency. These have resulted an increase of approximately 23 storms per year. While the year to year change is small, its effect when integrated over one half a century is marked! Further, the changes have not been linear, rather there are marked discontinuities in the time trend of the annual number of U.S. tornadoes. These discontinuities also are found when the storms are sorted by intensity. One striking feature is that while the total number of tornadoes reported each year is increasing, the yearly number of strong and violent tornadoes (those rated at F2 and greater) reports has actually decreased over the past half century.

The procedures used in the reporting of tornado attributes have also changed markedly. These changes include: the operational application of the F-scale during damage assessment; the growing reluctance to report of very long tornado tracks; and the criteria for reporting tornado track width. This is demonstrated through a comparison of the distributions of tornado attributes contained in the full data set to those produced using shorter data samples. The full data set will be used to illuminate features of the attribute distributions, and to explore the inter-relationships between track parameters. For instance it will be shown that track width is not a proxy for tornado intensity.

A conservative estimate of the nationwide distribution of tornado hazard is computed using the "Minimum Assumption Tornado-Hazard Model" (Schaefer et al., 1986). This distribution is compared to previously computed hazard estimates in order to estimate the sensitivity of the method to changes in the input data. If tornado risk is to be used for design and regulatory purposes, the computations need to be robust with respect to time.