However, as pointed out by Guillaumont and Guillaumont-Jeanneney (2006a), introducing squared and cubic aid terms could make it difficult to simultaneously treat possible conditional effects of aid. Furthermore, studies employing threshold models assume that the threshold is the same for all countries, which can be problematic given the very specific circumstances of recipient countries (Guillaumont & Guillaumont-Jeanneney, 2006a, 2006b). Moreover, on the interactive variable between foreign and institutional variables, Dalgaard and Hansen (2001) and Easterly (2003) highlighted that this specification does not allow disentangling the effects of aid and institutions on growth given that institutions themselves affect growth. Our paper proposes an innovative view by incorporating the possibility of heterogeneous effects of foreign aid in a general and unspecified way, and then exploring whether the quality of institutions is the source of such unobserved heterogeneity. We assume that there are multiple growth regimes and that the impact of foreign aid on growth differs across regimes. The approach allows us to study the role of governance by estimating whether the quality of governance affects the probability for a country to be in a given growth regime. Previous studies used traditional econometric models (OLS, IV, GMM) which impose a single model in the sample, and thus assume that the effect of foreign aid is constant across the distribution. These models also disregard the possibility that heterogeneity may exist along the distribution of the outcome itself. We employ a finite mixture model, which relaxes the assumption of a single model and allows unobserved heterogeneity in the sample. The finite mixture model incorporates a latent variable to classify countries into different classes or regimes, and enables any possible unobserved heterogeneity that may exist to be incorporated. In this semi-parametric model, countries are sorted into regimes or classes depending on the similarity of the conditional distribution of their growth rates given all the observed explanatory variables (Deb & Gregory, 2016; Konte, 2013). Finite mixture models are formulated to identify heterogeneous effects, if they exist, and characterise that heterogeneity along dimensions of the outcome distribution, observed characteristics and unobserved characteristics. These models are increasingly applied to different topics including health (