We explore the heterogeneous effect of migrant remittances on citizens' support for taxation using a sample comprising 45,000 individuals from the Afrobarometer survey round 7 [2016–2018] across 34 African countries. To correct for unobserved heterogeneity, we endogenously identify latent classes/subtypes of individuals that share similar patterns on how their support for taxation is affected by their unobserved and observed characteristics, including remittance dependency. We apply the finite multilevel mixture of regressions approach, a supervised machine learning method to detect hidden classes in the data without imposing a priori assumptions on class membership. Our data are best generated by an econometric model with two classes/subtypes of individuals. In class 1 where more than two-thirds of the citizens belong, we do not find any significant evidence that remittance dependence affects support for taxation. However, in class 2 where the remaining one-third of the citizens belong, we find a significant negative effect of remittance dependence on support for taxation. Furthermore, we find that citizens who have a positive appraisal of the quality of the public service delivery have a lower probability of belonging to the class in which depending on remittance reduces support for taxation. The findings emphasize the need for efficient public services provisioning to counteract the adverse effect of remittances on tax morale.