Commuting is the main source of CO₂ emissions from urban transport. However, existing studies have rarely paid attention to the differences in the effects of different geographical scales of built environments on residents' CO₂ emissions from commuting and had not yet reached a consensus conclusion. Based on the 2015 travel survey data and multilevel and mixed-effects models, this paper conducts an empirical study on the effects of multi-scale built environments on residents' CO₂ emissions from commuting in Guangzhou, China. The results show that after control for the residential self-selection effect, there are obvious spatial differences in the residents' CO₂ emissions from commuting among neighborhoods. It is shown that the residents in the central urban area generally emit less CO₂ emissions than their counterparts in the suburban area in commuting trips. These are caused by differences in built environments between neighborhoods. In terms of scale, the neighborhood's built environment has the most significant effect on residents' CO₂ emissions from commuting, followed by 1 km-buffer range of neighborhood boundary, and then subdistrict. Moreover, the effect of the built environment of the residence on CO₂ emissions from commuting is more significant than that of the workplace. These findings imply that planning interventions on the built environment should focus more on the neighborhoods in which residents live and the 15-minute walk life circle that is closely linked to the daily travel activities of residents. The distance between residence and workplace should be kept as short as possible, and the residential density of neighborhoods should be maintained at a reasonable level. Furthermore, optimizing the structure of road network and providing more community roads which are beneficial to non-motorized travel could help improve the environment for walking and bicycling and encourage people to use low-carbon, active and healthy travel modes. Although there may be some limitations in the selection of neighborhoods surveyed and random interception approach in the survey that may lead to non-possibility sampling, the conclusions can still provide a scientific basis for constructing a low-carbon urban spatial structure, guiding residents' travel behavior change and formulating targeted policies on low-carbon transportation and land use.