Previous studies have shown that introducing real principal component analysis (PCA) into wireless MIMO channel modeling can effectively improve its accuracy. However, wireless channel data is mostly in the form of complex-valued channel impulse response (CIR) and thus real PCA is very difficult to recover the phase of channel. Under such a motivation, this paper proposes a novel complex PCA-based MIMO channel modeling methodology. First, we prove that PCA can be used to analyze the measured complex-valued CIR data with the principle of maximizing the converted channel power. Then the representative channel features are extracted, which are utilized to reconstruct the CIR finally. An indoor measurement with 32 × 56 antennas is performed for validation. The results clearly show that the proposed channel model is more accurate compared to the real PCA-based model, and is tightly close to the measured channel. Moreover, the proposed methodology is robust with the increase of antennas, which provides insights into massive MIMO channel modeling in the future.