Abstract: A technique for automatically identifying clusters of data from a set of data samples, by employing multiple morphological operations of a grid of data cells representative of the data samples. The data samples are stored (12) in the data cell grid (56) as binary quantities of which the position in the grid represents each multivariate data point. Potential cluster regions are identified by first performing a series of dilation steps (22) on the data grid, to expand contiguous regions covered by the data points until smaller regions merge into larger ones that are identified as the potential cluster regions. Then a series of erosion operations (26) is performed, shrinking the contiguous regions until smaller isolated regions are completely eliminated and thin linkages between other potential cluster regions are removed. Next, a further series of dilation operations (30) expands the contiguous regions again, to reconnect any potential cluster regions that were fragmented in the erosion operations.