This paper is an extension of the well known rank order clustering algorithm for group technology problems. The ROC method is analysed and its main drawbacks are identified. The present method uses the ROC algorithm in conjunction with a block and slice method for obtaining a set of intersecting machine cells and non-intersecting part families. Then a hierarchical clustering method is applied based on a measure of association among pairs of machine cells. Clustering is terminated when all the surviving cells are non-intersecting or when a single group is formed. In the latter case, the number of cells is determined on the basis of a suitable decision criterion and the bottleneck machines are identified at the appropriate hierarchical level in the clustering process.