GOODS-

Context. Dust attenuation curves in external galaxies are useful for studying their dust properties as well as interpreting their intrinsic spectral energy distributions. These functions are not very well known in the UV range whether at low or high redshift. In particular, the presence or absence of a UV bump at 2175 Å remains an open issue that has consequences on the interpretation of broad band colours of galaxies involving the UV range. Aims. We want to study the dust attenuation curve in the UV range at z > 1 where the UV is redshifted into the visible and with Herschel data to constrain dust emission and a global dust attenuation. In particular, we search for a UV bump and related implications for dust attenuation determinations. Methods. We use deep photometric data of the Chandra Deep Field South obtained with intermediate and broad band filters by the MUSYC project to sample the UV rest frame of galaxies with 1 < z < 2. Herschel/PACS and Spitzer/MIPS data are used to measure the dust emission. We select 30 galaxies with high S/N in all bands. Their SEDs from the UV to the far-IR are fitted using the CIGALE code and the characteristics of the dust attenuation curves obtained as Bayesian outputs of the SED fitting process. Results. The mean dust attenuation curve we derive exhibits a significant UV bump at 2175 Å whose amplitude corresponds to 35% (76%) that of the Milky Way (Large Magellanic Cloud: LMC2 supershell) extinction curve. An analytical expression of the average attenuation curve (A(λ)/AV) is given and found to be slightly steeper than the Calzetti et al. one, although at a 1σ level. Our galaxy sample is used to study the derivation of the slopes of the UV continuum from broad band colours, including the rest frame GALEX FUV − NUV colour. Systematic errors induced by the presence of the bump are quantified. We compare dust attenuation factors measured with CIGALE to the slope of the UV continuum and find that there is a large scatter around the relation valid for local starbursts (~0.7 mag). The uncertainties on the determination of the UV slope lead to an extra systematic error in a range of 0.3 to 0.7 mag on dust attenuation when a filter overlaps the UV bump