We carried out a joint analysis of three grass removal experiments in which seedlings of various Fabaceae species were transplanted into plots with native grass and companion plots where grass had been removed. First, we estimated the effect of grasses on tree seedling mortality and seedling growth rate at each site. Then, we used the resulting coefficient estimates from site-level models to examine the impact of two climate (monthly precipitation and aridity index) and two soil (soil organic carbon content and clay content) variables on the direction and magnitude of the grass effects. Grasses increased the risk of mortality, but there was no evidence for a global effect of grasses on tree seedling rate of height growth. The best model fit indicated a high mortality risk of tree seedlings in response to grasses at intermediate aridity index values. No other climate or soil variable influenced tree seedling survival or growth (monthly precipitation, soil organic carbon content and clay content). Our results support the notion that the grass layer consistently creates a bottleneck to tree seedling establishment in African and South American savannas beyond climate and soil conditions, mainly by affecting tree seedling survival. The negative effect of grasses on seedling survival was lower in dry conditions compared to intermediate aridity levels. These results suggest that grass–seedling interaction is less intense in drier conditions, possibly due to reduced total grass biomass, which leads to decreased site evapotranspiration and improved soil water retention capacity.