Filamentary structures are an important part of star-forming interstellar clouds. The properties of filaments hold clues to their formation mechanisms and their role in the star-formation process. We compare the properties of filaments in the Orion Molecular Cloud 3 (OMC-3), as seen in mid-infrared (MIR) absorption and far-infrared (FIR) dust emission. We also wish to characterise some potential sources of systematic errors in filament studies. We calculated optical depth maps of the OMC-3 filaments based on the MIR absorption seen in Spitzer data and FIR dust emission observed with Herschel and the ArTeMiS instrument. We then compared the filament properties extracted from the data. Potential sources of error were investigated more generally with the help of radiative transfer models. The widths of the selected OMC-3 filament segments are in the range 0.03-0.1pc, with similar average values seen in both MIR and FIR analyses. Compared to the widths, the individual parameters of the fitted Plummer functions are much more uncertain. The asymptotic power-law index has typically values p~3 but with a large scatter. Modelling shows that the FIR observations can systematically overestimate the filament widths. The effect is potentially tens of per cent at column densities above N(H_2_)~10^22^cm^-2^ but is reduced in more intense radiation fields, such as the Orion region. Spatial variations in dust properties could cause errors of similar magnitude. In the MIR analysis, dust scattering should generally not be a significant factor, unless there are high-mass stars nearby or the dust MIR scattering efficiency is higher than in the tested dust models. Thermal MIR dust emission can be a more significant source of error, especially close to embedded sources. The analysis of interstellar filaments can be affected by several sources of systematic error, but mainly at high column densities and, in the case of FIR observations, in weak radiation fields. The widths of the OMC-3 filaments were consistent between the MIR and FIR analyses and did not reveal any systematic dependence on the angular resolution of the observations.