Given a spectral library, sparse unmixing aims to estimate the fractional proportions in each pixel of a hyperspectral image scene. However, the ever-growing dimensionality of spectral dictionaries strongly limits the performance of sparse unmixing algorithms. In this study, we propose a novel dictionary pruning (DP) approach to improve the performance of sparse unmixing algorithms, making them more accurate and time-efficient. We quantify the relative importance of each spectral dictionary atom using the total utility metric at virtually no cost. In this way, we have quantitative insights into how well the elements in the dictionary represent the hyperspectral scene. We evaluate the performance of the proposed dictionary pruning approach on several simulated data sets and one real data. We also compare the experimental results with two well-known dictionary pruning methods both visually and quantitatively and demonstrate the superiority of our proposed method through extensive experimental analysis.