Abstract Aim Species diversity is distributed heterogeneously through space, for reasons that are poorly understood. We tested three hypotheses to account for spatial variation in coniferous tree species diversity in a temperate island archipelago. The theory of island biogeography (ToIB) predicts that island area affects species diversity both directly (by increasing habitat diversity) and indirectly (by increasing abundances, which in turn reduce extinction rates). The ToIB also predicts that island isolation directly affects species diversity by reducing immigration rates. The passive sampling hypothesis predicts that island area and isolation both affect species diversity indirectly, by increasing and decreasing abundances, respectively. Community assembly rules (i.e. even partitioning of conifer abundances among islands) might also reduce tree species diversity beyond the core predictions of ToIB and the passive sampling hypothesis. Location Barkley Sound, British Columbia, Canada. Methods The abundances of eight coniferous tree species were quantified on 34 islands and two (1 ha) mainland plots. The predictions of the ToIB and the passive sampling hypothesis were tested using path analysis, and null models were used to test for abundance-based assembly rules and to further test the passive sampling hypothesis. Results Path analysis showed that island area and isolation did not have direct, statistical effects on tree species diversity. Instead, both geographic variables had direct statistical effects on total tree abundances, which in turn predicted tree diversity. Results from several passive sampling null models were correlated with observed patterns in species diversity, but they consistently overestimated the number of tree species inhabiting most islands. A different suite of null models showed support for community assembly rules, or that tree species often reached higher abundances on islands that housed fewer heterospecific trees. Main conclusions Results were inconsistent with the ToIB. Instead, patterns in tree diversity were best explained by a combination of stochastic (passive sampling) and deterministic (assembly rules) processes. Stochastic and deterministic processes are commonly considered to be exclusive explanations for island community structure, but results from this study suggest that they can work synergistically to structure island tree communities.