Areas of concentration of crime ("hot spots") have been characterized and analyzed by using various methods, such as point mapping, thematic boundary maps, and spatial ellipses. Recently, however, awareness of the limitations of these traditional methods gave rise to a new approach that creates smooth surface depiction of crime density based on point event maps. Among such new techniques, kernel density estimation is one of the most promising. In the current study, the crime data consisted of 20,158 penal code offense incidents known to the police during an 18-month period from January 1996 to June 1997 within the city. These data were address-geocoded with street-block precision onto a large-scale digital map. Using upgraded address-geocoding software, 19,462 incidents (96.5 percent) were successfully geocoded. Out of these, 1,572 incidents of burglary were drawn for analyses based on their spatial distribution. This was done by means of such spatial analysis software as CrimeStat and ArcView 3.0a Spatial Analyst Extension. The study found that a nearest neighbor analysis of burglary gave a nearest neighbor indicator value of 0.49, which indicates that the occurrence of burglary is not distributed randomly but tends to concentrate in certain locations. In applying kernel density estimation, a quartic function was selected as the kernel function, and the bandwidth was set to 250 m. The estimated density map indicated that 21.3 percent of all burglaries that occurred in the city were concentrated in areas that occupy no more than 4.5 percent of the whole city. When separate kernel density estimates were applied to residential burglaries and other burglaries, each of the two appeared to have its own "hot spots." These results imply that countermeasures focused on these "hot spots" may prove to be an efficient strategy for the reduction of the overall number of burglaries. 7 tables and 12 references