Land Surface Temperature (LST) is a markedly directional variable and its remotely sensed measurement may be strongly affected by viewing and illumination geometries. This study proposes the use of LST products collocated in space and time, but obtained with different viewing angles, to calibrate a simple model capable of characterizing the LST angular variability. The exercise is performed using MODIS (Aqua and Terra) and SEVIRI (Meteosat) LST products, for an area covering Mediterranean Europe and Northern Africa and encompassing the full years of 2011, 2012, 2013 and 2014. The approach relies on a kernel model that is composed by an "emissivity kernel" and a "solar kernel", associated to observation angle anisotropy and to shadowing/sunlit effects on the surface, respectively. The spatial distribution of the kernel coefficients is shown to reflect characteristics of the landscape, both in terms of vegetation cover and topography. Model performance is assessed through several comparison exercises over the 4-year period under analysis. Cross-validation results show that the angular correction by the kernel model leads to a decrease of the root mean square difference between SEVIRI and MODIS daytime (night-time) LST products, from the original uncorrected values of 3.5 K (1.5 K) to 2.3 K (1.3 K). Comparison of both MSG and MODIS LST products against in situ daytime measurements gathered over 2 years at a validation site in Évora (Portugal) reveals that the angular correction leads to a decrease in root mean square error from 4.6 K (2.0 K) to 3.8 K (1.9 K) for MODIS (SEVIRI). The kernel model may be a useful tool to quantify the LST uncertainties associated with viewing and illumination angles. Ermida et al., 2014, Duffour et al., 2015. This effect contributes to enhance the differences among LST satellite products, and therefore increasing the challenge of using multi-sensor and multi-decadal data to provide harmonised LST datasets suitable for long-term climate observations. Accurate estimates of the angular effects on retrieved LST are also crucial when performing in situ and cross-sensor validation exercises (Ermida et al., 2014). Quantification of these effects may also be relevant when using LST for model assessment (e.g. Wang et al. 2014; Trigo et al., 2015) and data assimilation (e.g. English 2008; Ghent et al., 2010).