Analysis of large-scale gene expression studies usually begins with gene clustering. A ubiquitous problem is that different algorithms applied to the same data inevitably give different results, and the differences are often substantial, involving a quarter or more of the genes analyzed. This raises a series of important but nettlesome questions: How are different clustering results related to each other and to the underlying data structure? Is one clustering objectively superior to another? Which differences, if any, are likely candidates to be biologically important? A systematic and quantitative way to address these questions is needed, together with an effective way to integrate and leverage expression results with other kinds of large-scale data and annotations. We developed a mathematical and computational framework to help quantify, compare, visualize and interactively mine clusterings. We show that by coupling confusion matrices with appropriate metrics (linear assignment and normalized mutual information scores), one can quantify and map differences between clusterings. A version of receiver operator characteristic analysis proved effective for quantifying and visualizing cluster quality and overlap. These methods, plus a flexible library of clustering algorithms, can be called from a new expandable set of software tools called CompClust 1.0 (http://woldlab.caltech.edu/compClust/). CompClust also makes it possible to relate expression clustering patterns to DNA sequence motif occurrences, protein-DNA interaction measurements and various kinds of functional annotations. Test analyses used yeast cell cycle data and revealed data structure not obvious under all algorithms. These results were then integrated with transcription motif and global protein-DNA interaction data to identify G1 regulatory modules.