What is the best way to reform Social Security? Academic literature offers diverging advice. There is a well-known result that the optimal size of Social Security is zero, implying it is best to phase the program out. Other studies argue that much can be gained by redesigning the program, given its current size. We provide a unified analysis that examines how the optimal size of Social Security depends on the key features of its design. We first develop a theoretical decomposition tracing the program's welfare effects to (i) income redistribution, (ii) distortions on the annuitization level, and (iii) intertemporal distortions. We then quantitatively assess the role of these channels. We show that the zero-optimal-size result arises because Social Security is too distortive and not redistributive enough. Once these design flaws are corrected, it is even optimal to increase the size of the program.