Determining the properties of starbursts requires spectral diagnostics of their ultraviolet radiation fields, to test whether very massive stars are present. We test several such diagnostics, using new models of line ratio behavior combining CLOUDY, Starburst99, and up-to-date spectral atlases. For six galaxies we obtain new measurements of He I 1.7 μm/Br10, a difficult to measure but physically simple (and therefore reliable) diagnostic. We obtain new measurements of He I 2.06 μm/Brγ in five galaxies. We find that He I 2.06 μm/Brγ and [O III]/Hβ are generally unreliable diagnostics in starbursts. The heteronuclear and homonuclear mid-infrared line ratios (notably [Ne III] 15.6 μm/[Ne II] 12.8 μm) consistently agree with each other and with He I 1.7 μm/Br10 this argues that the mid-infrared line ratios are reliable diagnostics of spectral hardness. In a sample of 27 starbursts, [Ne III]/[Ne II] is significantly lower than model predictions for a Salpeter initial mass function (IMF) extending to 100 Msolar. Plausible model alterations strengthen this conclusion. By contrast, the low-mass and low-metallicity galaxies II Zw 40 and NGC 5253 show relatively high neon line ratios, compatible with a Salpeter slope extending to at least ~40-60 Msolar. One solution for the low neon line ratios in the high-metallicity starbursts would be that they are deficient in >~40 Msolar stars compared to a Salpeter IMF. An alternative explanation, which we prefer, is that massive stars in high-metallicity starbursts spend much of their lives embedded within ultracompact H II regions that prevent the near- and mid-infrared nebular lines from forming and escaping. This hypothesis has important consequences for starburst modeling and interpretation.