There is currently much research interest in cancer prevention, particularly the role that dietary phytochemicals might play in blocking initiation of tumours or in delaying or reversing their progress. Thousands of in vitro studies in cell culture and many animal models representing a wide range of target tissues have shown anti-cancer efficacy for a large number of diet-derived compounds. However, success in clinical trials to date has been much more limited. Prevention trials in apparently healthy populations, with development of cancer as the main endpoint, require large numbers of individuals to be followed over many years, even decades, with enormous cost implications. To enable many more agents to be tested much more quickly, validated surrogate endpoint biomarkers are required, which will accurately determine outcome at a much earlier time in the process of tumour development. Such biomarkers are currently few and far between. One possibility is to identify at risk patients who have developed premalignant lesions, such as intraepithelial neoplasia, and monitor the progress of these in response to chemopreventive treatments. Several dietary phytochemicals, including indoles and polyphenols, have shown promise in this respect, with regression of respiratory papillomatosis, cervical and prostate intraepithelial neoplasia and oral leukoplakia. However, the relationship of some of these early lesions to tumour outcome is uncertain. Identification of reliable molecular markers would be advantageous to determine earlier changes, to relate directly to the carcinogenic process and, where possible, to allow less invasive assessment of efficacy. This requires a detailed knowledge not only of the stages of carcinogenesis for a particular tumour, but also of the mechanisms of action of the preventive agent. An ideal molecular marker would be closely linked to tumour development and treatment efficacy. Measurement directly or via a closely related activity should be possible, and preferably in non-invasive clinical specimens. Such biomarkers would be valuable in several aspects of chemoprevention – as targets to identify new agents or to optimise lead compounds; as risk biomarkers for selecting suitable cohorts for chemopreventive trials, or as indicators of efficacy for determining response to mechanism-based interventions or potential toxicity. There is no shortage of candidate proteins related to oncogenic processes (drug metabolising enzymes, growth factors, transcription factors, cell cycle and apoptosis related proteins) which have been shown to be modulated by phytochemicals in vitro. However, many of these mechanistic studies have been carried out with single high doses that are not achievable in vivo. Thus some of the reported effects may not be physiologically relevant. Effects can be cell type specific and so different panels of biomarkers may be required for different target tissues or for different cancer subtypes within a single tissue. On the other hand there appears to be a certain degree of similarity in the protein targets affected by a variety of structurally unrelated phytochemicals, suggesting similar mechanisms of action. A detailed understanding of the effects of dietary agents (for example on growth factor receptor signaling, epithelial to mesenchymal transition, cell cycle arrest and apoptosis) following extended treatment at physiologically achievable doses, with respect to agent, target tissue and cancer subtype, will help to identify useful biomarkers. Such an understanding would include identification of primary targets for phytochemicals (particular proteins such as receptors, or more general such as redox status), reasons why healthy cells are generally more resistant and comparison of in vitro with in vivo efficacy. Also of increasing importance is the investigation of combinations of phytochemicals, or their use in combination with other therapies, to increase efficacy or decrease unwanted side effects.