Cooperative networks recently became attractive because they can achieve spatial diversity. Multiple decode-and-forward (DF) relay nodes result in multiple carrier frequency offsets (CFOs) because each relay node owns its local oscillator. This paper presents a novel semi-blind multiple-CFO estimator for DF OFDM cooperative networks. A procedure is designed to effectively derive the semi-blind multiple-CFO estimator for relay nodes. Each relay node occupies its own subchannels, and its CFO falls in a nonoverlapped spectrum. The CFO mapping can simply match the corresponding relay nodes. In addition, the proposed method can reduce hardware and computational complexity. Semi-blind random sequences are derived from the characteristics of the signal matrix in the multiple signal characterization (MUSIC) algorithm. Comprehensive simulations show that random sequences can replace periodic training sequences. Furthermore, some information can be conveyed by random sequences to increase spectral efficiency.