Pulsed photothermal radiometry can be used for non-invasive depth profiling of optically scattering samples, including biological tissues such as human skin. Computational reconstruction of the laser-induced temperature profile from recorded radiometric signals is sensitive to the value of the tissue absorption coefficient in the infrared detection band (μ_IR). While assumed constant in reported reconstruction algorithms, μ_IR of human skin varies by two orders of magnitude in the commonly used 3−5 μm detection band. We analyse the problem of selecting the effective absorption coefficient value to be used with such algorithms. In a numerical simulation of photothermal profiling we demonstrate that results can be markedly impaired, unless the reconstruction algorithm is augmented by accounting for spectral variation μ_IR(λ). Alternatively, narrowing the detection band to 4.5–5 μm reduces the spectral variation μ_IR(λ) to a level that permits the use of the simpler, un-augmented algorithm. Implementation of the latter approach for depth profiling of port wine stain birthmarks in vivo is presented and discussed.