The aim of this work is to identify HeII emitters at 2<z<4.6 and to constrain the source of the hard ionizing continuum that powers the HeII emission. We have assembled a sample of 277 galaxies with a high quality spectroscopic redshift at 2<z<4.6 from the VVDS survey, and we have identified 39 HeII1640A emitters. We study their spectral properties, measuring the fluxes, equivalent widths (EW) and FWHM for most relevant lines. About 10% of galaxies at z~3 show HeII in emission, with rest frame equivalent widths EW0~1-7A, equally distributed between galaxies with Lya in emission or in absorption. We find 11 high-quality HeII emitters with unresolved HeII line (FWHM_0<1200km/s), 13 high-quality emitters with broad He II emission (FWHM_0>1200km/s), 3 AGN, and an additional 12 possible HeII emitters. The properties of the individual broad emitters are in agreement with expectations from a W-R model. On the contrary, the properties of the narrow emitters are not compatible with such model, neither with predictions of gravitational cooling radiation produced by gas accretion. Rather, we find that the EW of the narrow HeII line emitters are in agreement with expectations for a PopIII star formation, if the episode of star formation is continuous, and we calculate that a PopIII SFR of 0.1-10 Mo yr-1 only is enough to sustain the observed HeII flux. We conclude that narrow HeII emitters are either powered by the ionizing flux from a stellar population rare at z~0 but much more common at z~3, or by PopIII star formation. As proposed by Tornatore et al. (2007), incomplete ISM mixing may leave some small pockets of pristine gas at the periphery of galaxies from which PopIII may form, even down to z~2 or lower. If this interpretation is correct, we measure at z~3 a SFRD in PopIII stars of 10^6Mo yr^-1 Mpc^-3 qualitatively comparable to the value predicted by Tornatore et al. (2007).