The concept of neighborhood remains important in criminology, in particular in relation to the social disorganization school of thought and its subsequent development into collective efficacy theory. Most neighborhood based studies make use of administrative geographical units such as census tracts although it is widely recognized as problematic due to the potential mismatch between the boundaries of administrative geographical units and boundaries of real importance to people or mechanisms of interest. This is called the Modifiable Areal Unit Problem (MAUP), which consist both of problems of scale (size of neighborhoods) and of zonation (boundary placement). In the present study data over arson from the Swedish rescue services 2007-2012 have been employed to analyze MAUP in the city of Malmö. It is argued that arson is a form of physical disorder that is strongly related to social disorganization. The analysis is based on a comparison of administrative geographical units of analysis compared to randomly generated geographical units of similar size. In the analysis both SAMS-areas (n=391) and part-areas (n=136) are considered alongside a total of 40 sets of randomly generated geographical units of similar size. The city has been divided into 50*50 meter micro-places (n=64544) which have been assigned a value for arson based on incidence, kernel density and distance to nearest arson. Empty two-level hierarchical linear regression models with the micro-places as level 1 unit have been employed to calculate Intra-Class Correlations (ICC) separately for each of the 42 different geographical units of analysis as level 2 units. Results show that administrative geographical units of analysis in many cases just are marginally better than geographical units with random borders if the basic urban structure is taken into account. Implications for further research is discussed.