Using a sample of college graduates from the NLSY97, we introduce a new approach to assessing wage benefits of STEM training, STEM jobs, and the match between the two: rather than classify individuals dichotomously as STEM or non-STEM, we measure the STEM-intensities of both their college coursework and their occupational requirements. While the orthodox approach simply predicts that 'STEM pays,' we find that workers at the top of both gender-specific STEM-intensity distributions are predicted to out-earn their counterparts at the bottom by a substantial margin -- even when we condition on their dichotomous STEM classification -- but that predicted log-wages do "not" increase monotonically with STEM-intensity throughout the entire joint distribution.