We present a method for selecting high-redshift Type la supernovae (SNe la) located via rolling SN searches. The technique, using both color and magnitude information of events from only two to three epochs of multiband real-time photometry, is able to discriminate between SNe la and core-collapse SNe. Furthermore, for SNe la the method accurately predicts the redshift, phase, and light-curve parameterization of these events based only on pre-maximum-light data. We demonstrate the effectiveness of the technique on a simulated survey of SNe la and core-collapse SNe, where the selection method effectively rejects most core-collapse SNe while retaining SNe la. We also apply the selection code to real-time data acquired as part of the Canada-France-Hawaii Telescope Supernova Legacy Survey (SNLS). During the period 2004 May to 2005 January in the SNLS, 440 SN candidates were discovered, of which 70 were confirmed spectroscopically as SNe la and 15 as core-collapse events. For this test data set, the selection technique correctly identifies 100% of the identified SNe II as non-SNe la with only a 1%-2% false rejection rate. The predicted parameterization of the SNe la has a precision of | Δz|/(1 + z spec) < 0.09 in redshift and ±2-3 rest-frame days in phase, providing invaluable information for planning spectroscopic follow-up observations. We also investigate any bias introduced by this selection method on the ability of surveys such as SNLS to measure cosmological parameters (e.g., w and Ω M) and find any effect to be negligible. © 2006. The American Astronomical Society. All rights reserved.