Abstract In 1977 an experiment was initiated in the Mojave Desert to investigate the relationship between shrub interactions and structure in a community dominated by Ambrosia dumosa and Larrea tridentata. Here, as in much of the Mojave, Larrea were regularly distributed, Ambrosia occurred in aggregations, and the two were randomly distributed relative to each other. Pre‐dawn xylem pressure potentials (PDXPPs) of single Ambrosia or Larrea in centres of 100 m2 circular plots were monitored to assess effects of intraspecific, interspecific and total removals of neighbouring shrubs. Contrary with theory, results over the next 2 years indicated that interspecific interference was more intense than intraspecific interference in both species. These plots were maintained through 2014. Measurements of seedling recruitment from 1980 to 2014, and of PDXPP, above‐ground biomass, and canopy senescence from 2003 to 2014 were conducted. Recruitment of both species was substantial immediately after the removals, but declined to very low levels after 1983. Ambrosia recruited into all Ambrosia and Total‐removal plots, but Larrea recruited only into plots that contained mature Ambrosia. Pre‐dawn xylem pressure potentials of monitored shrubs continued to be enhanced in removal plots for at least 27 years, but this changed from most being due to interspecific removals in both species to intraspecific removals causing most enhancement in Ambrosia and inter‐ and intraspecific removals causing nearly equal enhancements in Larrea. Above‐ground biomasses of monitored Ambrosia and Larrea were 2.1× and 2.8× larger in Total‐removal plots, 1.6× and 1.7× larger in intraspecific removal plots, respectively, and 1.1× larger in interspecific removal plots for both species than those in Control plots, indicating the absence of intraspecific interference had the dominant long‐term effect. Canopy senescence differed between Ambrosia and Larrea in extent, timing and effect of specific removal treatments; it was greatest for both species in Controls, averaging 75% and 34%, respectively. Synthesis. Shrub interactions and their relations to community structure are mechanistically and spatially complex. Differences between short‐term and long‐term responses to removals reveal multi‐tiered, temporally dynamic feedback loops between shrub interactions and community structure driven by demographics, species‐specific root growth, resource competition, communications and territoriality.