Existing algorithms, including the stochastic point‐source method, used to simulate synthetic ground‐motion records are aimed at sampling records at a single station or records at multiple stations. The application of the algorithms may not adequately reproduce the observed coherency structure of the actual records and the intraevent spatial correlation characteristics of peak ground acceleration or spectral accelerations. To improve these, we suggest an extension to the stochastic point‐source method by introducing a target spatial coherency structure and the spatially correlated uncertainties in the Fourier amplitude spectrum for each recording station. The use of the extended model to simulate the multiple‐station records is illustrated, and the spatial correlation of the ground‐motion measures for the simulated records are compared with the empirical spatial correlation model derived based on the 1999 Chi‐Chi Taiwan earthquake.