Database reverse engineering (DBRE) attempts to recover the technical and semantic specifications of he persistent data of information systems. Dependencies between records (data dependency) form a major class that need to be recovered. Since most of these dependencies are not supported by the DBMS, (foreign keys are the main exception, at least in modern relational DBMS), they have not be explicitly declared in the database schema. Careless reverse engineering will inevitably ignore them, leading to poor quality conceptual schema. Several information sources can contribute to the elicitation of these hidden dependencies. The program source code has long been considered the richest, but also the most complex, of them. In this paper, we analyze and compare, through their respective quality and cost, different program understanding techniques that can be used to elicit data dependencies.