Abstract. The spatial join operation is both one of the most important and expensive operations in Geographic Database Management Systems (GDBMS). This paper presents a set of rules to optimize the performance of the filtering step of spatial joins operations. First, a set of expressions to predict the number of I/O operations and CPU performance is presented. The rules are based on expressions to predict the performance of algorithms and tests performed with synthetic and real data sets. Fos some cases, the optimized algorithm can execute the same operation 10 times faster than the original, non-optimized version.