We show that using confidence-weighted classification in transition-based parsing gives results comparable to using SVMs with faster training and parsing time. We also compare with other online learning algorithms and investigate the effect of pruning features when using confidenceweighted classification.