We decompose firm size into four components: the lagged 5-year component that represents size five years ago, and the long-run, intermediate-run, and short-run components that capture changes in size in each horizon. Our analyses indicate that while the lagged 5-year component explains about 80% of the cross-sectional variation in size, it has little return predictability. In contrast, the long-run change in size component explains only 18% of size, but it completely captures the size premium. Our decomposition also sheds light on the January effect, the disappearance of the size premium since early 1980s, and the return behaviors of new entrants.