In this critical literature review, we seek to understand why multidimensional, psychological measures of human emotion that have been popular in the study of emotion and learning to date, may not yield the statistical power or construct validity necessary to consistently explain or predict human learning. We compare competing theories and conclude that educational studies tend towards use of multi-dimensional models of human emotions which, while useful in educational psychology and therapeutic practice, suffer from psychometric flaws and generate lower power when used as empirical research constructs compared with the "basic emotion" models and their derivatives popular in the neurobiological, cognitive, and social sciences. Based on our review, we conclude that more extensive use of physiological measures and analysis of spontaneous emotion language, both rooted in the basic emotions tradition rather than continued psychological measurement of multi-dimensional emotions, may yield more consistent and significant results and reduce education researchers' reliance on self-report measures. Findings from the review may advance the selection of operational definitions and formulation of research questions for new empirical studies of the intersections between emotion and learning.