<p>Scientists often rely on proxies when they identify or measure new or complex phenomena. However, these tools are frequently seen as epistemologically inferior because they are indirect and make it difficult to properly control for confounding factors. This view implies two methodological norms. First, if possible, proxies should be replaced with more direct and better-controlled tools. Second, if proxies cannot be replaced, they should be improved by increasing control over confounding factors. We evaluate this view by examining functional genomics, a field in which proxies are abundant. Analysing work from the largest contemporary initiative in functional genomics (ENCODE), we observe that researchers do not follow these methodological norms. Instead, they continue using the same proxies and even combine them to produce novel insights or measurements. This potentially paradoxical finding can be explained if we recognise that proxies have a “generativity” that is appreciated and leveraged by the researchers working with them. Proxies in functional genomics form a dynamic toolkit that equips researchers to discover unexpected insights and ask new questions. Our analysis thereby contributes to a better understanding of the epistemic roles that proxies play in scientific practice.</p>