Modern data-intensive software systems manipulate an increasing amount of data in order to support users in various execution contexts. Maintaining and evolving activities of such systems rely on an accurate documentation of their behavior which is often missing or outdated. Unfortunately, standard program analysis techniques are not always suitable for extracting the behavior of data-intensive systems which rely on more and more dynamic data access mechanisms which mainly consist in run-time interactions with a database. This paper proposes a framework to extract behavioral models from data-intensive program executions. The framework makes use of dynamic analysis techniques to capture and analyze SQL execution traces. It applies clustering techniques to identify data manipulation functions from such traces. Process mining techniques are then used to synthesize behavioral models.