High-accuracy models are essential for rapid performance estimations and reliable parameter refinements during the preliminary design of centrifugal compressors. In this paper, a hybrid performance prediction method is proposed by combining the loss models of the single-zone method and the division idea of the two-zone method. Physically-based loss models are adopted for the jet zone analysis in substitute of the empirical parameters in the two-zone method. Two additional parameters are introduced for the wake zone analysis to implicitly evaluate the loss not fully understood in the single-zone method. For these two-zone parameters, prediction models are proposed by first analyzing the dominant factors from one-dimensional level and then extracting the coefficients from diverse datasets with statistical techniques. The hybrid method is validated on four centrifugal compressors with flow coefficients ranging from 0.09 to 0.58 and pressure ratios up to 8.5. The results show that, compared with the single-zone and two-zone methods, the hybrid method provides more reliable performance predictions with higher accuracy in the whole operating range.