AbstractUsing transient imaging data from the 2^nd and 3^rd years of the SDSS supernova survey, we apply various machine learning techniques to the problem of classifying transients (e.g. SNe) from artefacts, one of the first steps in any transient detection pipeline, and one that is often still carried out by human scanners. Using features mostly obtained from PCA, we show that we can match human levels of classification success, and find that a K-nearest neighbours algorithm and SkyNet perform best, while the Naive Bayes, SVM and minimum error classifier have performances varying from slightly to significantly worse.