Accurately projecting hydrological responses in climate-sensitive arid regions is a critical scientific challenge under global warming. This study investigates summer precipitation in the arid region of northwest China (ANWC) from the perspective of synoptic regime frequency. Using a hidden Markov model, we objectively identify dominant summer rainfall patterns over ANWC and their long-term variations in occurrence frequencies from 1961 to 2022. The underlying physical mechanism for each rainfall pattern is then elucidated through circulation composites and Lagrangian water vapor tracking. We further examine and explain the model biases in summer precipitation over ANWC based on occurrence frequencies of rainfall patterns. Finally, the potential bias in future changes that stems from the inaccurate historical baseline is analyzed. Results show that summer precipitation over ANWC is dominated by a dry pattern, but its occurrence frequency has significantly decreased. CMIP6 models exhibit a historical dry bias primarily due to the systematic overestimation in the occurrence frequency of the dominant dry pattern. Thus, the systematic biases likely underestimate the magnitude of future increase in summer precipitation. Our key finding moves beyond simply identifying a mean-state error to pinpoint a specific error in the simulated frequency of a physically coherent weather regime.