Prediction of target detection performance is the basis for radar adaptive resource scheduling and refined information processing design. However， conventional statistical models are difficult to accurately describe the statistical distribution of the radar scattering cross section of stealth targets， resulting in inaccurate detection performance prediction. For this reason， this paper proposes a radar detection performance evaluation method based on a hybrid chi-square distribution model， which models and evaluates the detection probability under constant false alarm conditions. Based on the hybrid model， the target detection probability is modeled under the condition of constant false alarm using the folded line approximation and Kummer transform method； the accuracy of the model is examined by Monte Carlo simulation； the hybrid model is compared with the traditional statistical model from the perspective of detection probability by fixing the false alarm rate. The simulation results show that compared with the traditional Swerling model， the single-pulse detection performance of the hybrid cardinality model has a large difference in different azimuth ranges， which better reflects the real detection scenarios and provides theoretical support for the overall optimization design of the stealth target detection system.