Fragmentation effects on biodiversity remain contentious, with landscape-scale studies reporting positive effects on species richness while fragment-level studies document diversity losses. We tested whether fragmentation per se affects taxonomic and functional diversity independently and whether landscape heterogeneity mediates these effects. We surveyed small mammal communities across 131 Atlantic Forest sites in Brazil, embedded in landscapes varying in habitat amount, fragmentation, compositional heterogeneity, and matrix quality. We quantified taxonomic richness and four functional diversity metrics based on ten morphological, behavioral, and dietary traits, using hierarchical approaches to test fragmentation effects across six spatial scales (500-5,500 m) and structural equation models to evaluate mediation by landscape heterogeneity. Fragmentation per se increased species richness independent of habitat amount, with effects detectable only at landscape scales (R² = 19% at 5,500 m; β = 0.27, p < 0.01). However, fragmentation simultaneously reduced functional dispersion (β = −0.27, p < 0.01) without affecting standardized functional richness, revealing complete taxonomic-functional decoupling. Community-weighted mean traits revealed that fragmentation favored smaller species with reduced dispersal capacity (body mass: β = −0.35, p < 0.01; dispersal: β = −0.25, p < 0.05). Landscape heterogeneity did not mediate these effects, rejecting the niche diversification hypothesis. Positive taxonomic effects of fragmentation per se mask functional homogenization driven by accumulation of disturbance-tolerant generalists while functionally unique species are filtered out. Conservation strategies relying solely on species inventories systematically overestimate ecological resilience in fragmented landscapes. Understanding fragmentation’s true impact requires functional assessment beyond species counts, fundamentally changing how we evaluate and conserve biodiversity in humanity’s increasingly fragmented world.