The current study explored the score gap in reading achievement between the Flemish community and the French community in Belgium based on the Programme for International Student Assessment (PISA) 2009 database. Four student background variables, two school composition variables, four school resources variables, five school climate variables, and four school governance variables were used as predictors. The multilevel analyses indicated that the effect of each predictor might vary between communities. The Oaxaca-Blinder decomposition of the community level score gap in mean indicated that each community might benefit more from the endowments and/or the returns of different predictors. The Juhn-Murphy-Pierce decomposition of the community level score gap in percentiles indicated that some predictors might be explanatory for the between-community differences among lower-achievers, with some others being explanatory for differences among higher-achievers. It would be valuable to school improvement, if education policy-makers and practitioners could consider to implement those “significant” predictors specifically influential in their community. Meanwhile, the overall improvement of some predictors at community level might be beneficial to its average achievement. Simultaneously, the selective manipulation of some predictors for lower-, medium-, or higher-achievers within one specific community might be beneficial.