Using input and outcome data from a randomized study of dual-language immersion programs in an urban district, we examine the mediating relationships of dosage, expenditures, and classroom characteristics to students' academic performance, and the moderating role of students' race/ethnicity. Differential costs of immersion were concentrated at the district level and were modest, at about 2% to 4% of per-pupil spending annually. We estimate that an additional US$100 spent per immersion student in a given year was associated with an additional 8% of a standard deviation in language arts performance in English, which was just over one third of the causal point-in-time enrollment effect of 22% of a standard deviation. We find no generalizable evidence of differential effects by race/ethnicity.