This chapter addresses the challenge of model order reduction for linear systems subject to non-probabilistic uncertainties. A novel convex set-based reduced-order model (CSBROM) framework is established to effectively handle parameter uncertainties with unknown-but-bounded (UBB) and parameter correlation of uncertainties using convex set theory rather than large-sample statistical methods. The approach begins by reconstructing the uncertain system into a convex set-based (CSB) state-space equation via an order-extended matrix technique, ensuring both controllability and stability. Subsequently, the classical balanced truncation method is generalized to the convex domain. A convex set perturbation-based singular value decomposition (SVD) algorithm is derived to accurately estimate the bounds of uncertain Hankel singular values (HSVs) by solving the CSB Lyapunov equation. To determine the optimal dimension of the reduced system, a novel truncation criterion is proposed based on the geometric possibility of uncertain singular value distributions. Numerical examples confirm that the proposed CSBROM achieves accuracy comparable to Monte Carlo simulations (MCSs) with significantly higher computational efficiency.