Large samples of high-redshift supernovae (SNe) are potentially powerful probes of cosmic star formation, metal enrichment and SN physics. We present initial results from a new deep SN survey, based on re-imaging in the R, i', z' bands, of the 0.25deg^2^ Subaru Deep Field (SDF), with the 8.2-m Subaru telescope and Suprime-Cam. In a single new epoch consisting of two nights of observations, we have discovered 33 candidate SNe, down to a z'-band magnitude of 26.3 (AB). We have measured the photometric redshifts of the SN host galaxies, obtained Keck spectroscopic redshifts for 17 of the host galaxies and classified the SNe using the Bayesian photometric algorithm of Poznanski et al. (2007AJ....134.1285P) that relies on template matching.