Large areas of tropical forest have been cleared and planted with exotic grass species for use as cattle pasture. These often remain persistent grasslands after grazer removal, which is problematic for restoring native forest communities. It is often hoped that remnant and/or planted trees can jump-start forest succession; however, there is little mechanistic information on how different canopy species affect community trajectories. To investigate this, I surveyed understory communities, exotic grass biomass, standing litter pools, and soil properties under two dominant canopy trees—Metrosideros polymorpha (‘ōhi‘a) and Acacia koa (koa)—in recovering Hawaiian forests. I then used structural equation models (SEMs) to elucidate direct and indirect effects of trees on native understory. Native understory communities developed under ‘ōhi‘a, which had larger standing litter pools, lower soil nitrogen, and lower exotic grass biomass than koa. This pattern was variable, potentially due to historical site differences and/or distance to intact forest. Koa, in contrast, showed little understory development. Instead, data suggest that increased soil nitrogen under koa leads to high grass biomass that stalls native recruitment. SEMs suggested that indirect effects of trees via litter and soils were as or more important than direct effects for determining native cover. It is suggested that diverse plantings which incorporate species that have high carbon to nitrogen ratios may help ameliorate the negative indirect effects of koa on natural understory regeneration.