In an analysis of learners' reflections taken in real time, this study attempts to unveil what learners attend to during a task session. A total of 110 Japanese learners of English noted rationales for their form selections of epistemic modal verbs. The data were then coded and tallied to enable quantitative analysis and to examine the association of individual rationales with the accuracy of the selected forms. In-depth analyses of representative reflections are also presented in a qualitative inquiry. Among the total counts of some 1400 reflections, a number of patterns emerged, including textual-meaning-oriented rationales, meta-linguistic terminology use, and modal-meaning-oriented rationales. A combination of qualitative and quantitative analyses of each category of rationales indicated highly significant results for some and insignificant results for others. Overall outcomes indicate that when learners' attention is oriented to the form-meaning connection in context, learners become successful in comprehension. Further, the potential of formulaic learning is discussed. (Contains 5 notes and 5 tables.)