The growth and repair of normal tissues depends on a small sub-population of cells termed somatic stem cells whose primary characteristic is an ability for indefinite self-renewal. Epithelial stem cells divide to produce cells, termed transient amplifying cells, that undergo a few rounds of more rapid division before they terminally differentiate. Evidence that the growth of tumours, as for normal tissues, is ultimately dependent on a subpopulation of the proliferatively competent cells was first shown for leukaemias by isolation of small sub-populations of phenotypically distinct 'tumour-initiating cells'. Differing cell surface phenotypes also prospectively identify tumour-initiating sub-populations in solid tumours. Even cell lines derived from tumours retain hierarchical stem cell patterns demonstrable as differing clonogenic abilities related to cellular properties such as size, adhesiveness, dye exclusion, and patterns of gene expression. Malignant stem cells appear to form the primary targets of therapy, but how differences between malignant stem and other cells affect therapeutic responses remains unclear. However, transplantation methods exist for their analysis and the in vitro persistence of stem cell patterns may provide systems for developing new therapeutic approaches.