Emission line galaxies (ELGs) are used in several ongoing and upcoming surveys (SDSS-IV/eBOSS, DESI) as tracers of the dark matter distribution. Using a new galaxy formation model, we explore the characteristics of [OII] emitters, which dominate optical ELG selections at $zsimeq 1$. Model [OII] emitters at $0.5<z<1.5$ are selected to mimic the DEEP2, VVDS, eBOSS and DESI surveys. The luminosity functions of model [OII] emitters are in reasonable agreement with observations. The selected [OII] emitters are hosted by haloes with $M_{rm halo}geq 10^{10.3}h^{-1}{rm M}_{odot}$, with ~90% of them being central star-forming galaxies. The predicted mean halo occupation distributions of [OII] emitters has a shape typical of that inferred for star-forming galaxies, with the contribution from central galaxies, $langle N rangle_{left[OIIright], cen}$, being far from the canonical step function. The $langle N rangle_{left[OIIright], cen}$ can be described as the sum of an asymmetric Gaussian for disks and a step function for spheroids, which plateaus below unity. The model [OII] emitters have a clustering bias close to unity, which is below the expectations for eBOSS and DESI ELGs. At $zsim 1$, a comparison with observed g-band selected galaxy, which are expected to be dominated by [OII] emitters, indicates that our model produces too few [OII] emitters that are satellite galaxies. This suggests the need to revise our modelling of hot gas stripping in satellite galaxies.