Identifying the factors controlling stream water pollutants is challenged by the diversity of potential sources, pathways, and processes. This study tests the effects of watershed characteristics on stream water metal concentrations across environmental gradients. By using an extensive data set of 83 watersheds in southern Finland and structural equation modeling (SEM), the direct and indirect effects of land use and soil type on metal concentrations were explored. Both land use and soil type resulted in statistically significant direct effects on metals; for example, land use was found to control dissolved metal concentrations, whereas soil type had the strongest links for total metal concentrations. The consideration of indirect correlation further strengthened the effects of soil type up to 50%, thus suggesting the dominant role of soil across land use intensities. Moreover, the results indicate that modified landscapes mediate the effect of natural soil processes in controlling stream metal concentrations. This work highlights the benefits of structural equation model framework, as the underlying paths for water quality are more likely to be identified, compared to traditional regression methods. Thus, the implementation of SEM on water quality studies is highly encouraged.