Tumor initiation (TI) in xenotransplantation models of head and neck squamous cell carcinoma (HNSCC) is an inefficient process. Poor TI could be due to (1) posttransplant cell loss, (2) a rare sub-population of cancer stem cells or (3) a requirement for specific cellular interactions, which rely on cell number. By tracking GFP-expressing HNSCC cells, we conclude that the posttransplant loss of cancer cells is minimal in the xenotransplant model. Furthermore, an examination of putative cancer stem cell markers (such as CD133, CD44, SP and label retention) in HNSCC cell lines revealed no correlation between marker expression and tumorigenicity. In addition, single-cell clones randomly isolated from HNSCC cell lines and then transplanted into mice were all capable of initiating tumors with efficiencies varying almost 34-fold. As the observed variation in the clones was both more and less tumorigenic than the parental cells, a combination of two clones, at suboptimal cell numbers for TI, was implanted into mice and was found to modulate the tumor-initiating activity, thus indicating that TI is dependent on a 'critical' number of cells and, for the first time, that interactions between clonal variants within tumors can modulate the overall tumor-initiating activity. Put in context with previous literature on tumorigenic activity, we believe that interactions between clonal variants within a tumor as well as (1) stromal interactions, (2) angiogenic activity, (3) immunocompetence and (4) cancer stem cells may all contribute to tumorigenic potential and the propensity for tumor growth and recurrence.