A series of dot and choropleth maps showed the clustering of both incendiary and accidental fires not only in specific Springfield neighborhoods but also along individual city streets. To identify the varying frequency of arson fires among the 36 census tracts in the study area, multiple regression analysis was performed using census data reflecting a variety of social, economic, and housing characteristics. Two variables representing housing vacancy and tenement-type housing entered into the final regression equation. Together these two variables accounted for 70.4 percent of the reported arsons. A subsequent inspection of the regression residuals revealed a random geographical pattern, thereby precluding any simple explanation for the remaining unexplained arson. Factors such as insurance, building ownership, and length of occupancy warrant analysis in future research. 7 figures, 3 tables, and 18 references. (Author abstract modified)