We present the results of the first observations of the emission line galaxies (ELG) of the extended Baryon Oscillation Spectroscopic Survey. From the total 9000 targets, 4600 have been selected from the Dark Energy Survey (DES). In this subsample, the total success rate for redshifts between 0.6 and 1.2 is 71 and 68 per cent for a bright and a faint samples, respectively, including redshifts measured from a single strong emission line. The mean redshift is 0.80 for the bright and 0.87 for the faint sample, while the percentage of unknown redshifts is 15 and 13 per cent, respectively. In both cases, the star contamination is lower than 2 per cent. We evaluate how well the ELG redshifts are measured using the target selection photometry and validating with the spectroscopic redshifts measured by eBOSS. We explore different techniques to reduce the photometric redshift outliers fraction with a comparison between the template fitting, the neural networks and the random forest methods. Finally, we study the clustering properties of the DES SVA1 ELG samples. We select only the most secure spectroscopic redshift in the redshift range 0.6 < z < 1.2, leading to a mean redshift for the bright and faint sample of 0.85 and 0.90, respectively. We measure the projected angular correlation function and obtain a galaxy bias averaging on scales from 1 to 10 Mpc h(-1) of 1.58 +/- 0.10 for the bright sample and 1.65 +/- 0.12 for the faint sample. These values are representative of a galaxy population with M-B - log(h) < -20.5, in agreement with what we measure by fitting galaxy templates to the photometric data.