This article describes a method to estimate the time lags between stations of a dense strong-motion seismography array and to obtain composite (as opposed to frequency-dependent) spatial correlation functions. The main features of the proposed procedure are the closure property, expressing the need for internal consistency in the pattern of lags, and the use of relative maxima of cross-correlation functions, yielding multiple estimates of each lag and the associated correlation coefficient. Results obtained by processing 12 events recorded by the SMART1 array provide insight into the propagation of seismic waves across a dense-array site, enabling the development of a regression model for the lags. The simplest model assumes plane waves traveling at constant velocity in the epicentral direction, and the lag estimation yields both the local-mean propagation velocity and a set of inferred deviations from this local mean. Results are also presented for the spatial correlation of the horizontal components of the aligned accelerations for the 12 SMART1 events, and the decay of composite correlation with distance is tentatively interpreted in terms of the frequency content of the ground motions.