The determination of oxygen abundance in nebulae requires measuring a significant number of emission lines distributed along a wide spectral range. The required measurements are hard to obtain at high redshift, where sources are very faint, and where the accessible spectral range is limited. These difficulties are often overcome using empirical relationships between the oxygen abundance and the fluxes in a small number of strong lines. The so-called strong-line methods are often the only practical alternative for metallicity estimate at high redshift. In this sense, the low metallicities range is particularly important since high redshift objects are primitive and so of low metallic content. One of the most widely used relationships links the oxygen with the ratio between [NII]6583 and Hα. This relationship shows a large scatter at low metallicity. In an effort to bring down the errors, we re-calibrated the relationship using a large sample of extremely metal-poor galaxies. The SDSS spectra of the galaxies were all analyzed in the same way to minimize systematic errors. To our surprise, the decrease of scatter reveals that the ratio [N{II}]6583 to Hα seems to be independent of metallicity at low oxygen abundance (12+log[{O}/{H}] < 7.6). This result casts doubts on the metallicities of high-redshift objects based on the relationship. We explain how the re-calibration was carried (including the sample selection and the abundance determinations). In addition, we try explain what produces the lack of correlation.