The classification scheme proposed by Walborn et al. (2002AJ....123.2754W
), based primarily on the relative strengths of the NIVλ4058 and NIIIλ4640 emission lines, has been used in a variety of studies to spectroscopically classify early O-type stars. Owing to the lack of a solid theoretical basis, this scheme has not yet been universally accepted though. We provide first theoretical predictions for the NIVλ4058/NIIIλ4640 emission line ratio in dependence of various parameters, and confront these predictions with results from the analysis of a sample of early-type LMC/SMC O-stars. Stellar and wind parameters of our sample stars are determined by line profile fitting of hydrogen, helium and nitrogen lines, exploiting the helium and nitrogen ionization balance. Corresponding synthetic spectra are calculated by means of the NLTE atmosphere/spectrum synthesis code fastwind. Though there is a monotonic relationship between the NIV/NIII emission line ratio and the effective temperature, all other parameters being equal, theoretical predictions indicate additional dependencies on surface gravity, mass-loss, metallicity, and, particularly, nitrogen abundance. For a given line ratio (i.e., spectral type), more enriched objects should be typically hotter. These basic predictions are confirmed by results from the alternative model atmosphere code cmfgen. The effective temperatures for the earliest O-stars, inferred from the nitrogen ionization balance, are partly considerably hotter than indicated by previous studies. Consistent with earlier results, effective temperatures increase from supergiants to dwarfs for all spectral types in the LMC. The relation between observed NIVλ4058/NIIIλ4640 emission line ratio and effective temperature, for a given luminosity class, turned out to be quite monotonic for our sample stars, and to be fairly consistent with our model predictions. The scatter within a spectral sub-type is mainly produced by abundance effects. Our findings suggest that the Walborn et al. (2002AJ....123.2754W
) classification scheme is able to provide a meaningful relation between spectral type and effective temperature, as long as it is possible to discriminate for the luminosity class. In terms of spectral morphology, this might be difficult to achieve in low-Z environments such as the SMC, owing to rather low wind-strengths. According to our predictions, the major bias of the classification scheme is due to nitrogen content, and the overall spectral type-Teff
relation for low-metallicity (e.g., SMC) O-stars might be non-monotonic around O3.5/O4.