Cancer therapy has been revolutionised by the advent of targeted agents. As a result, patient selection through genetic testing is a rapidly growing industry, but this has posed extra challenges for the patient and oncological team including cost, legal obstacles, test accuracy, and management of tissue. In this context, a recent report in the New England Journal of Medicine suggests intra-tumoral heterogeneity might be an even more complex issue than previously thought. The report showed that up to two-thirds of genetic results in renal-cell carcinoma will vary throughout the tumour specimen when tissue is assessed in a multiregional sequencing analysis. So what are the consequences of this finding for the use of single biopsies to guide treatments in the clinic? One of the first targeted drugs to show efficacy was imatinib for patients with chronic myeloid leukaemia. Imatinib exploits the BCR-ABL translocation, and has improved 5-year survival to about 90%. Since imatinib, there has been an explosion of new agents culminating most recently in vemurafenib and ipilimumab for patients with melanoma. The key to the success of most targeted therapies is the selection of patients who will respond—so-called patient enrichment—which means testing for the proposed target. Without selection, a dilution of the therapeutic effect occurs where a new treatment might be abandoned because patients without the target, and who do not respond, mask the activity in those that do. The reverse situation, testing to exclude patients from treatment, can also apply—eg, patients with colorectal cancer and KRAS mutations in exons 12 and 13 do poorly with EGFR inhibitors. With the growing complexity of tests, targets, and treatment pathways, obtaining tissue sufficient to meet therapeutic needs has become ever more important. Tissue might need to be sampled at different times, as therapeutic targets can change over time, and samples must also be of appropriate quality and representative of the tumour as a whole. Although intra-tumoral heterogeneity has long been appreciated, the NEJM report is alarming because it suggests that heterogeneity could seriously hamper accurate patient selection, and not just the therapeutic ratio—as a result, patients could be inappropriately treated with expensive drugs they are unlikely to respond to, or be omitted from active treatment. This is particularly clear in cancers for which taking multiple biopsies or surgical samples can be challenging. Lung cancer illustrates this point. The EGFR inhibitors, gefitinib and erlotinib, are well-established agents for patients with non-small-cell lung cancer. Most patients who respond to these inhibitors have activating mutations in exons 19 or 21 of EGFR, whereas those without such mutations do not. Selection is therefore vital to avoid dilution of efficacy and reduce patient harm. Furthermore, according to speakers at the 13th European Congress on Perspectives in Lung Cancer held in Amsterdam, Netherlands, on March 9–10, 2012, the role of the pathologist is paramount in ensuring not only that enough tissue is left after histological diagnosis to do essential tests—such as those to predict benefit from the EGFR inhibitors—but also that testing is not done until the evidence linking the target to the efficacy of the drug, and any potential future treatments, are considered. A multidisciplinary environment is therefore essential to ensure an appropriate discussion of the timing of any pathological testing, and, where possible, to enable tissue to be saved to help guide future decision-making. This point was affirmed in a recent comment in the Journal of Clinical Oncology in which the authors suggested that excluding patients without the proposed target for treatment of breast and gynaecological cancer with PI3K/mTOR inhibitors should not be done until “a stronger predictive relationship emerges”. The comment also acknowledged the challenge of testing for rare mutations. Resistance to therapy can also develop, necessitating further testing to select potential responders to subsequent therapy. Use of blood might help overcome some of the difficulties associated with tissue resources, and several studies have been done using serum samples to define predictors of response to treatment, but these investigations are fraught with difficulties, including linking any serum changes to the tumour itself. Deciding who to test, how, and when, will help realise the vast potential of personalised medicine for cancer. However, integration of biology-driven therapy into the clinic is becoming increasingly complex and intratumoral heterogeneity could prove to be the Achilles' heel of targeted therapy in an era in which tissue has become an extremely precious resource that must be used sparingly, pragmatically, and within a multidisciplinary decision-making team.