Context. Very massive stars (VMS) have masses in excess of 100 M ⊙ and are rare. However, owing to their powerful winds, very high luminosity, and efficient nucleosynthesis, they are key players of star-forming regions. In particular, their strong ionizing fluxes impact the surrounding interstellar medium. Aims. We aim at detecting VMS in local star-forming region from the imprint they leave on the integrated UV and optical light. Methods. We analyzed a sample of 27 star-forming regions and galaxies in the local Universe. We selected sources with a metallicity close to 12+log(O/H) = 8.3, which is typical of the Large Magellanic Cloud. We defined empirical criteria to distinguish sources dominated by VMS and Wolf-Rayet stars (WR), using template spectra of VMS- and WR-dominated regions. We subsequently built population synthesis models with an updated treatment of VMS. We investigated the successes and failures of these new models in accounting for the UV-optical spectroscopy of our sample sources. Results. We show that the UV range alone is not sufficient to distinguish between VMS- and WR-dominated sources because their spectra are almost identical in this range. The region of the WR bumps in the optical breaks the degeneracy. In particular, the morphology of the blue bump at 4640–4686 Å is a key diagnostic. Beyond the prototypical R136 region, which contains VMS, we identify two galaxies showing clear signatures of VMS. In two other galaxies or regions the presence of VMS can be suspected, as already discussed in the literature. The stellar population is clearly dominated by WR stars in seven other sources. The most recent BPASS population synthesis models can neither account for the strong He II 1640 emission, nor for the shape of the blue bump in VMS- and WR-dominated sources. Our models that include VMS more realistically reproduce the UV-optical spectra of VMS-dominated sources. Conclusions. We conclude that VMS are present in some local star-forming regions, but that separating them from WR-dominated populations requires optical spectroscopy with a high signal-to-noise ratio. A high equivalent width of He II 1640 is not a sufficient condition for identifying VMS. Populations synthesis models need to take VMS into account by incorporating not only evolutionary tracks, but also dedicated spectral libraries. Finally, we stress that the treatment of WR stars needs to be improved as well.