The Ecosystem Management Emulating Natural Disturbances (EMEND) project tests the hypothesis that varying levels of green tree retention maintain and retain forest biodiversity better than conventional clear-cutting. We studied epigaeic spiders to assess biodiversity changes two, five and ten years following a range of partial retention harvests (clear-cut, 10-75% retention) and unharvested controls in four boreal mixedwood cover-types. A total of 56, 371 adult spiders representing 220 species was collected using pitfall traps. Lasting effects on forest structure were proportional to harvest intensity. These changes strongly influenced spider richness, abundance and species composition, as well as assemblage recovery. Distinctive assemblages were associated with disturbance level, especially with partial harvests (≤50% retention), and these were dominated by open-habitat species even ten years after harvest. Assemblages were more similar to those of controls in the highest (75%) retention treatment, but significant recovery toward the structure of pre-disturbance assemblages was not detected for any prescription in any cover-type. Although early responses to retention harvest suggested positive effects on spider assemblages, these are better explained as lag effects after harvest because assemblages were less similar to those of unharvested controls five years post-harvest, and only minor recovery was observed ten years following harvest. Retention of forest biodiversity decreased over time, especially in conifer stands and the lower (10-50%) retention treatments. Overall, retention harvests retained biodiversity and promoted landscape heterogeneity somewhat better than clear-cutting; however, there was a clear gradient of response and no retention ‘threshold’ for conservation can be recommended on the basis of our data. Furthermore, results suggest that retention harvest prescriptions should be adjusted for cover-type. We show that low retention ameliorated impacts in broadleaved forests characteristic of earlier stages in mixedwood succession, but only higher retention was associated with less impact in successionally older conifer forests. Although these short-term responses (10 years) of spider assemblages support use of retention harvests, understanding the true conservation merit of these practices, relative to conventional approaches, requires evaluation over longer time scales, with work more focused on recovery of biodiversity than on its preservation after harvest.