Bacterial behavior and physiology during human infection is difficult to study and largely unknown, as our vast knowledge of infection microbiology is primarily derived from studies using in vitro and animal models. A key challenge to assessing bacterial physiology during human infection is the difficulty in acquiring and assessing bacterial function in human-derived samples. Here, I will discuss the use of microbial metatranscriptomics from chronic human wound, lung, and oral infections to tackle this gap in knowledge. We have leveraged these data in two primary ways: to assess and improve the accuracy of pre-clinical infection models using a quantitative framework recently developed in our lab; and identifying and functionally characterizing genes of unknown function that are highly expressed in humans but not in most pre-clinical models. I will also discuss additional approaches we are using to quantify microbial biogeography and heterogeneity within human infections, with the goal of using these data to develop accurate pre-clinical models for antimicrobial discovery.