We present new techniques for improving the efficiency of supernova (SN) classification at high redshift using 64 candidates observed at Gemini North and South during the first year of the Supernova Legacy Survey (SNLS). The SNLS is an ongoing 5 year project with the goal of measuring the equation of state of dark energy by discovering and following over 700 high-redshift SNe Ia using data from the Canada-France-Hawaii Telescope Legacy Survey. We achieve an improvement in the SN Ia spectroscopic confirmation rate: at Gemini 71% of candidates are now confirmed as SNe Ia, compared to 54% using the methods of previous surveys. This is despite the comparatively high redshift of this sample, in which the median SN Ia redshift is z = 0.81 (0.155 , z , 1.01). These improvements were realized because we use the unprecedented color coverage and light curve sampling of the SNLS to predict whether a candidate is a SN Ia and to estimate its redshift, before obtaining a spectrum, using a new technique called the "SN photo-z." In addition, we have improved techniques for galaxy subtraction and SN template j super(2) fitting, allowing us to identify candidates even when they are only 15% as bright as the host galaxy. The largest impediment to SN identification is found to be host galaxy contamination of the spectrum - when the SN was at least as bright as the underlying host galaxy the target was identified more than 90% of the time. However, even SNe in bright host galaxies can be easily identified in good seeing conditions. When the image quality was better than 0.55, the candidate was identified 88% of the time. Over the 5 year course of the survey, using the selection techniques presented here, we will be able to add 6170 more confirmed SNe Ia than would be possible using previous methods.