Eastern deciduous forests are changing in species composition and diversity outside of classical successional trajectories. Three disturbance mechanisms appear central to this phenomenon: fire frequency is reduced, canopy gaps are smaller, and browsers are more abundant. Which factor is most responsible is a matter of great debate and remains unclear, at least partly because few studies have simultaneously investigated more than one process. We conducted a large-scale experiment in mesophytic forests of West Virginia, USA, to test three key hypotheses: (1) the fire hypothesis (fire suppression limits diversity to few shade-tolerant, fire-intolerant species that replace and suppress many fire-tolerant species); (2) the gap hypothesis (small gaps typical of today's forests promote dominance of a few shade-tolerant species); and (3) the browsing hypothesis (overbrowsing by deer limits diversity to a few unpalatable species). We tested these hypotheses using a factorial experiment that manipulated surface fire, large canopy gap formation (gap size 255 m2), and browsing by deer, and we followed the fates of >28 000 seedlings and saplings for five years. Understory tree communities in control plots were dominated (up to 90%) by Fagus grandifolia, averaging little more than two species, whereas overstories were diverse, with 10–15 species. Fire, large canopy gaps, and browsing all dramatically affected understory composition. However, our findings challenge views that fire and large canopy gaps can maintain or promote diversity, because browsers reduced the benefits of gaps and created depauperate understories following fire. Consequently, two major disturbances that once promoted tree diversity no longer do so because of browsing. Our findings appear to reconcile equivocal views on the role of fire and gaps. If browsers are abundant, these two disturbances either depress diversity or are less effective. Alternatively, with browsers absent, these disturbances promote diversity (three- to fivefold). Our results apply to large portions of eastern North America where deer are overabundant, and we provide compelling experimental evidence that historical disturbance regimes in combination with low browsing regimes typical of pre-European settlement forests could maintain high tree species diversity. However, restoring disturbances without controlling browsing may be counterproductive.