Supernova Refsdal, multiply imaged by cluster MACSJ1149.5+2223, represents a rare opportunity to make a true blind test of model predictions in extragalactic astronomy, on a time scale that is short compared to a human lifetime. In order to take advantage of this event, we produced seven gravitational lens models with five independent methods, based on Hubble Space Telescope (HST) Hubble Frontier Field images, along with extensive spectroscopic follow-up observations by HST, the Very Large and the Keck Telescopes. We compare the model predictions and show that they agree reasonably well with the measured time delays and magnification ratios between the known images, even though these quantities were not used as input. This agreement is encouraging, considering that the models only provide statistical uncertainties, and do not include additional sources of uncertainties such as structure along the line of sight, cosmology, and the mass sheet degeneracy. We then present the model predictions for the other appearances of SN Refsdal. A future image will reach its peak in the first half of 2016, while another image appeared between 1994 and 2004. The past image would have been too faint to be detected in existing archival images. The future image should be approximately one third as bright as the brightest known image (i.e., H_AB~25.7 mag at peak and H_AB~26.7 mag six months before peak), and thus detectable in single-orbit HST images. We will find out soon whether our predictions are correct.