Abstract: To design a measurement system used in the large-scale sea area for passive underwater acoustic localization, the understanding about accuracy distribution characteristics is required. Aiming at this problem, a distributed accuracy estimation method was presented based on a bearings-only localization system with multi-stations. The factors of solution procedure, accuracy analysis and observation geometry design were integrated considered. The nonlinear least squares method was adopted to establish the solving model, then the random error propagation algorithm was proposed base on Monte-Carlo method, and the accuracy distribution described by the root mean square error(RMSE) was derived by large sampling and gridding simulation. The simulation tests proved that this model had quick convergence and reasonable response to the measurement factor change. In the case of a 30km×20km area covered by a 5-elements star-structure array, the RMSE median achieved 1km as the azimuth error of single station changed from 2° to 5°. The localization performance under different observation geometry conditions were comparable based on these accuracy distribution characteristics. The simulation results showed that the pentagon-structure array provided a more optimal accuracy distribution than the star-structure, rectangle-structure and trapezoid-structure array with a 5-elements array, which made the RMSE median reduce 0.1km and the high-accuracy covering area expand.. This method provides a technical way for accuracy analysis and observation geometry optimization in measurement system design.