The units of analysis were the 68 cities within the continental United States with available homicide data and a minimum population of 250,000 in 2001. Homicide counts and population for these cities were obtained from the Uniform Crime Reports for the years 1979 to 2001. Tobit regressions were used for all cities at risk of experiencing a cycle in order to estimate unbiased effects of theoretically important predictors on the timing of the phase changes for homicide. The findings indicate that the homicide epidemic of the 1980s and 1990s was widespread throughout urban America, suggesting that homicide patterns were not driven by the disproportionate impact of a small number of "mega-cities." Also, the timing of the advent and downturn was not distributed randomly throughout the period. The distribution of the dates for the respective phases of the homicide cycle shows a relatively small number of cities in the vanguard. In addition, the timing of the phase changes was not symmetrical, and it was apparently related to population structure and levels of deprivation. Among cities exhibiting this pattern, homicide rates were likely to begin to increase and then decline significantly earlier in larger, denser cities and in those with more extreme socioeconomic deprivation. There was some evidence of a geographic patterning of the timing of phase changes. The findings provide some support for Blumstein's and Rosenfeld's theory that the homicide epidemic correlated with the crack cocaine epidemic among Black urban males; however, the interplay of factors is complex, and their spatial patterning has yet to be fully explained. Further inquiry into homicide factors is required. 2 tables, 5 figures, and 45 references