This article analyzes the out-of-sample performance of portfolio optimization models in the US corporate bond universe. In our empirical study, we measure the benefits of naive diversification and find that it eventually reaches a limit as the number of bonds increases. Also, we observe substantial improvements in the risk-adjusted performance of scientific portfolio constructions when compared to simple barbell strategies for the same given duration. When duration constraints are relaxed, we find that both naively and scientifically diversified portfolios outperform cap-weighted benchmarks in terms of Sharpe ratio.