The psychophysical experimental results were simulated by a simple model based on the autocorrelation function (Priestly, 1981) operating on the image intensity values. The model uses a directional autocorrelation function, comparing vertical and horizontal autocorrelations. Since the autocorrelation function operates on the image intensity values (pixels), no assumptions are made on featural differences between the local line elements that give rise to similarity grouping (e.g. X vs k). Departing from classical autocorrelation models of vision (Uttal, 1975), we suggest that the interactions underlying the correlation measurements are time dependent in a way that long-range correlations take more time to be effective, while short range correlations can be detected within a relatively short processinlg time. This hypothesis is implemented in the model by multiplying the autocorrelation function l:,y an exponential weighting function with a time dependent space constant.