This study aims to evaluate the effects of urban geometry on retrieval of emissivity and surface temperature in urban areas. An improved urban emissivity model based on sky view factor (IUEM-SVF) was further enhanced to consider all radiance contributions leaving the urban canopy, including (i) emission by all facets within an instantaneous field of view (IFOV); (ii) reflection by all facets of emission from surrounding facets; and (iii) propagation of emitted and reflected radiation with multiple reflections (scattering) within a complex 3D array of urban objects. The effective emissivity derived from IUEM-SVF was evaluated with a microscale radiative transfer and energy balance model: Temperatures of Urban Facets in 3-D (TUF-3D). IUEM-SVF performs well when urban facets have uniform emissivity and temperature; e.g., root mean square deviations (RMSD) are less than 0.005 when material emissivity is larger than 0.80 (ɛ ⩾ 0.80). However, when material emissivities are variable within the observed target, differences of effective emissivity between IUEM-SVF and TUF-3D become larger, e.g., RMSD of 0.010. When the effect of geometry is not considered and a mixed pixel emissivity is defined, the difference is even much larger (i.e. 0.02) and this difference increases with the decrease of sky view factor. Thus, the geometry effect should be considered in the determination of effective emissivity. Effective emissivity derived from IUEM-SVF was used to retrieve urban surface temperature from a nighttime ASTER thermal infrared image. Promising results were achieved in comparison with standard LST products retrieved with the Temperature and Emissivity Separation (TES) algorithm. IUEM-SVF shows promise as a means to improve the accuracy of urban surface temperature retrieval. The effect of thermal heterogeneity on the effective emissivity was also evaluated by TUF-3D, and results show that the thermal heterogeneity cannot be neglected since the RMSD between the effective emissivity based on TUF-3D and IUEM-SVF is relatively large.