Taking advantage of the unprecedented combination of sensitivity and angular resolution afforded by the Herschel Space Observatory at far‐infrared and submillimetre wavelengths, we aim to characterize the physical properties of cold dust within nearby galaxies, as well as the associated uncertainties, namely the robustness of the parameters we derive using different modified blackbody models. For a pilot subsample of the KINGFISH (Key Insights on Nearby Galaxies: A Far‐Infrared Survey with Herschel) key programme, we perform two‐temperature fits of the Spitzer and Herschel photometric data (from 24 to 500 μm), with a warm and a cold component, both globally and in each resolution element. We compare the results obtained from different analysis strategies. At global scale, we observe a range of values of the modified blackbody fit parameters βc (0.8–2.5) and Tc (19.1–25.1 K). We compute maps of our modelling parameters with βc fixed or treated as a free parameter to test the robustness of the temperature and dust surface density maps we deduce. When the emissivity is fixed, we observe steeper temperature gradients as a function of radius than when it is allowed to vary. When the emissivity is fitted as a free parameter, barred galaxies tend to have uniform fitted emissivities. Gathering the parameters obtained in each resolution element in a Tc–βc diagram underlines an anticorrelation between the two parameters. It remains difficult to assess whether the dominant effect is the physics of dust grains, noise, or mixing along the line of sight and in the beam. We finally observe in both cases that the dust column density peaks in central regions of galaxies and bar‐ends (coinciding with molecular gas density enhancements usually found in these locations). We also quantify how the total dust mass varies with our assumptions about the emissivity index as well as the influence of the wavelength coverage used in the fits. We show that modified blackbody fits using a shallow emissivity (β < 2.0) lead to significantly lower dust masses compared to the β < 2.0 case, with dust masses lower by up to 50 per cent if βc = 1.5, for instance. The working resolution affects our total dust mass estimates: masses increase from global fits to spatially resolved fits.