Prompted by international concerns about school and postschool participation in mathematics, the authors of the present study sought to investigate factors predicting "switching on" and "switching off" in mathematics, operationalized through measures of future intent and disengagement, respectively. They drew from Bronfenbrenner's ecological model to explore an educational ecology comprising student, home, classroom, school, and time factors in predicting future intent and disengagement. These models were conducted within a multilevel context, thereby also assessing variation at student, classroom, and school levels. Based on data from 1,601 Australian middle school students from 200 classrooms in 44 schools, variance components models showed the bulk of variance (greater than 85%) in future intent and disengagement resided at the student (and residual) level, with 10%-12% variance explained at classroom and school levels. Multilevel hierarchical regression analyses revealed that student, home, and classroom factors explained most variance in dependent measures. For future intent and disengagement, mathematics self-efficacy, valuing, enjoyment, perceived classroom enjoyment, and parent interest were significant predictors. For disengagement, additional predictors were mathematics anxiety, perceived classroom disengagement, school ethnic composition, and school socioeconomic status. Modeling indicated all predictive effects generalized across classrooms and schools. Findings hold implications for the breadth and depth of efforts in mathematics aimed at reducing disengagement, increasing participation, and promoting more positive future orientations through school and beyond. (Contains 2 tables, 5 footnotes, and 2 figures.)