Key uncertainties remain in accurately measuring soil respiration, including how the commonly-used technique of collar insertion affects measured soil and root-derived CO2 fluxes. We hypothesized that total soil respiration is frequently under-estimated because soil collar insertions sever surface roots, which coupled with the preferential practice of taking daytime measurements, leads to the autotrophic (root-derived) component frequently being missed. We measured root distribution and soil CO2 efflux in three contrasting ecosystems: a Lodgepole pine (Pinus contorta) plantation, an upland heather-dominated peatland and a lowland sheep-grazed grassland, where we combined shallow surface collars with collars at different soil insertion depths for occasional and continuous hourly flux measurements. Collar insertion by only a few centimetres reduced total soil CO2 efflux in all three ecosystems by an average of 15% but at times by up to 30–50%, and was directly proportional to the quantity of cut fine roots. Most reduction occurred in the shallow-rooted peatland system and least in the deep-rooted grassland. In the forest and grassland, soil temperatures explained most of the deep-collar (largely heterotrophic) variation and did not relate to the root-derived (largely autotrophic) flux component, whilst the opposite was true for the peatland site. For the forest, the autotrophic flux component peaked at night during moist periods and was drought-limited. Mean flux estimates differed between sampling time and insertion depth. Our results suggest strongly that accurate measurement and modelling of soil respiration needs explicitly to consider collar insertion, and the root-derived flux component, with its own temperature sensitivity and potential time-lag effects.