As anisotropy of land surface thermal emissivity (8~14μm) reduces the accuracy of the remotely sensed land surface temperature (LST) and its application, which is more important for urban surface, a kernel model was suggested in this paper to describe the urban surface emissivity anisotropy (USEA). The USEA is estimated by the ratio between off-nadir and nadir observed emissivity. The kernel model has two basic assumptions: (1) at daytime, USEA has apparent hot spot whose location approaches the sun’s location;(2) at nighttime, there is no apparent hot spot effect in USEA, but it has significant correlation with viewing zenith angle and weak correlation with azimuth angle. Three kernels named isotropic kernel, multi-scattering kernel, and temperature difference kernel constitute the kernel-based model, where the isotropic kernel is a constant equal to 1, the multi-scattering kernel describes the dependence on viewing zenith angle, and the temperature difference kernel describes the hot spot effect of USEA. Results, based on computer simulations, indicated that the kernel model can accurately describe the spatiotemporal variation of USEA but the thermal inertia of urban surface would reduce the applicability of the kernel model. The kernel model has many potential applications with MODIS directional land surface emissivity data.