We studied canopy gaps over a range of stand ages and site moisture classes in the temperate rain forest of Clayoquot Sound, southwestern British Columbia, Canada. We tested predictions about the landscape mosiac at three levels of resolution: the forest as a whole, gaps, and gapmakers. In addition to canopy gaps arising from the patchy mortality of dominant trees (developmental gaps), we described the prevalence and characteristics of gaps arising from edaphic and topographic features. Roughly 56% of the forest ares is influenced to some degree by canopy openings (canopy gap + expanded gap), with 30% of forest area in canopy gaps averaged over all sites (14% developmental canopy gaps and 16% in edaphic canopy gaps). Within the most intensively sampled biogeoclimatic subzone, 73% of gaps were developmental in origin. The majority of edapthic gaps result from streamcourses. Though percent area in developmental canopy gaps was similar across stand age classes, gaps and gapmaker were more common in old growth than in mature stands. Most developmental gaps resulted from the mortality of more than one tree: 96% of the developmental gaps had more than one gapmaker, and 50% had three or more. Mature stands had a significantly higher number of gapmakers per gap than old—growth stands. Old—growth forests were thus dominated by many small gaps and mature forests by fewer larger gaps. Drier and wetter sites had more gapmakers per gap than did stands on mesic sites. The number of decay classes of gapmakers in developmental gaps increases with number of gapmakers per gap, indicating that larger gaps represent the combined effects of distinct mortality events separated in time, rather than single large events. Gapmakers in developmental gaps were distributed broadly and evenly over the range of decay classes, suggesting a continuous pattern of recruitment of gapmakers rather than larger episodic mortality events. The snapping of boles and standing death were the most common modes of gapmakers mortality, together accounting for 76% of the mortality of gapmakers. Uprooting was the least common mode of gapmaker mortality in both mature and old—growth forests (23.6% and 15.6%, respectively; 20.6% of all gapmakers) and uprooting was more common in drier and wetter sites than in mesic sites. These general trends in gap—phase structure and gap formation processes are consistent with data on gaps and forests age structure collected elsewhere on the British Columbia coast, suggesting that a regime of small—scale, low—intensity disturbance is common in the old—growth forests of coastal British Columbia. We estimate that, in the absence of large—scale disturbances, turnover time for this forest is between 350 and 950 yr, resulting in ubiquitous late successional character throughout the landscape mosaic. Silvicultural practices intended to emulate the natural disturbance regime of these forests should create small gaps of 3—10 trees in an otherwise continuous forest matrix and should retain substantial late—successional characteristics within managed forests.