In many regions of the world, variable retention has replaced clear-cutlogging as the principal method of regeneration harvest. Partial retention of the overstory is thought to ensure greater continuity of the species and ecological processes that characterize older forests. Level (amount) and spatial pattern of overstory retention are two basic elements of forest structure that can be manipulated to achieve specific ecological or silvicultural objectives. However, experiments that elucidate the relative importance of retention level and pattern (or their interaction) are rare. Here we assess long-term (>10 yr) responses of forest understories to experimental harvests of mature coniferous forests replicated at five sites in the Pacific Northwest (PNW). Treatments contrast both the level of retention (40% vs. 15% of original basal area) and its spatial distribution (dispersed vs. aggregated in 1-ha patches). For most vascular plant groups (early seral, forest generalist, and late seral), postharvest changes in cover and richness were reduced at higher levels of retention and in dispersed relative to aggregated treatments. Although retained forest patches were stable, changes in adjacent harvested (cleared) areas were significantly greater than in dispersed treatments. Late-seral herbs were highly sensitive to level and pattern of retention, with extirpations most frequent in the cleared areas of aggregated treatments and at low levels of dispersed retention. In contrast, early-seral species were most abundant in these environments. Forest-floor bryophytes exhibited large and persistent declines regardless of treatment, suggesting that threshold levels of disturbance or stress were exceeded. Our results indicate that 15% retention (the minimum standard on federal forestlands in the PNW) is insufficient to retain the abundance or diversity of species characteristic of late-seral forests. Although 1-ha aggregates provide refugia, they are susceptible to edge effects or stochastic processes; thus, smaller aggregates are unlikely to serve this function. The ability to achieve multiple ecological or silvicultural objectives with variable retention will require the spatial partitioning of habitats to include dispersed retention and larger undisturbed aggregates along with cleared areas.