Biological processes often manifest themselves as coordinated changes across modules, i.e., sets of interacting genes. Commonly, the high dimensionality of genome-scale data prevents the visual identification of such modules, and straightforward computational search through a set of known pathways is a limited approach. Therefore, tools for the data-driven, computational, identification of modules in gene interaction networks have become popular components of visualization and visual analytics workflows. However, many such tools are known to result in modules that are large, and therefore hard to interpret biologically. Here, we show that the empirically known tendency towards large modules can be attributed to a statistical bias present in many module identification tools, and discuss possible remedies from a mathematical perspective. In the current absence of a straightforward practical solution, we outline our view of best practices for the use of the existing tools.