We test state-of-the-art model atmospheres for young very-low-mass stars and brown dwarfs in the infrared, by comparing the predicted synthetic photometry over 1.2–24 μm to the observed photometry of M-type spectral templates in star-forming regions. We find that (1) in both early and late young M types, the model atmospheres imply effective temperatures (${{T}_{{\rm eff}}}$) several hundred Kelvin lower than predicted by the standard pre-main sequence (PMS) spectral type–${{T}_{{\rm eff}}}$ conversion scale (based on theoretical evolutionary models). It is only in the mid-M types that the two temperature estimates agree. (2) The ${{T}_{{\rm eff}}}$ discrepancy in the early M types (corresponding to stellar masses $\gtrsim 0.4$ ${{M}_{\odot }}$ at ages of a few Myr) probably arises from remaining uncertainties in the treatment of atmospheric convection within the atmospheric models, whereas in the late M types it is likely due to an underestimation of dust opacity. (3) The empirical and model-atmosphere J-band bolometric corrections are both roughly flat, and similar to each other, over the M-type ${{T}_{{\rm eff}}}$ range. Thus the model atmospheres yield reasonably accurate bolometric luminosities (${{L}_{{\rm bol}}}$), but lead to underestimations of mass and age relative to evolutionary expectations (especially in the late M types) due to lower ${{T}_{{\rm eff}}}$. We demonstrate this for a large sample of young Cha I and Taurus sources. (4) The trends in the atmospheric model J−Ks colors, and their deviations from the data, are similar at PMS and main sequence ages, suggesting that the model dust opacity errors we postulate here for young ages also apply at field ages.