In this paper, regression-based and clustering-based index tracking methods are compared in terms of tracking accuracy, solution consistency, portfolio volatility, and downside risk. The former is based on least-squares regression under a cardinality constraint. The latter is based on K-means, K-medoids, and hierarchical clustering algorithms with dissimilarity metrics defined on Euclidean distance, Pearson correlation coefficient, and dynamic time warping. Experimental results on major world stock markets show that the regression-based method significantly outperforms the clustering-based methods in terms of tracking accuracy and consistency, while the index tracking method based on hierarchical clustering and Pearson correlation coefficient results in slightly lower volatility and downside risk than the regression-based method.