The restoration of disturbed ecosystems is challenging and often unsuccessful, particularly when non-native plants are abundant. Ecosystem restoration may be hindered by the effects of non-native plants on soil biogeochemical characteristics and microbial communities that persist even after plants are removed. To examine the importance of soil legacy effects, we used experimental restorations of Florida shrubland habitat that had been degraded by the introduction of non-native grasses coupled with either mechanical disturbance or pasture conversion. We removed non-native grasses and inoculated soils with native microbial communities at each degraded site, then examined how habitat structure, soil nitrogen, soil microbial abundances, and native seed germination responded over two years compared to undisturbed native sites. Grass removal treatments effectively restored some aspects of native habitat structure, including decreased exotic grass cover, increased bare ground, and reduced litter cover. Soil fungal abundance was also somewhat restored by grass removals, but soil algal abundance was unaffected. In addition, grass removal and microbial inoculation improved seed germination rates in degraded sites, but these remained quite low compared to native sites. High soil nitrogen persisted throughout the experiment regardless of treatment. Many treatment effects were site-specific, however, with legacies in the more degraded vegetation type tending to be more difficult to overcome. These results support the need for context-dependent restoration approaches and suggest that the degree of soil legacy effects may be a good indicator of restoration potential.