Hierarchical clustering procedures have received a great deal of emphasis in recent years, yet research has lagged in their empirical evaluation and in objective means to aid the user in selecting good partitions (rather than good hierarchies). The present paper aims to correct both of these deficiencies, first by empirically testing selected methods which have become popular and, second by proposing and evaluating statistical stopping rules. Results indicate: 1. that methods vary widely in their performance and 2. that the proposed stopping rules can aid the user in selecting partitions.