This work proposes a domain-decomposition approach for the least-squares Petrov–Galerkin (LSPG) reduced-order model. The DD-LSPG method enables reduced-order models to be constructed independently on subdomains of the computational domain, which facilitates parallel computation and enables localized basis adaptation. The method employs a nonoverlapping domain decomposition with interface conditions that ensure global consistency. Numerical experiments demonstrate that the DD-LSPG method can achieve accuracy comparable to global LSPG while enabling computational parallelism and flexibility in basis construction.