The guiding principle of early work was the hypothesis that the anti-cancer alkylating drugs acted through their ability to cross-link macromolecules essential for cell division. Not long afterwards, DNA was specified as the essential target, and support for the hypothesis came from evidence that the archetypal agent, mustard gas, could link guanine bases in DNA through their N-7 atoms. Quantitative correlations between alkylation of DNA and its inactivation as a template followed, with bacteriophage as a simple test object, showing that the mean lethal dose was close to a single cross-link in the genome. This conclusion applied to either mustard gas or the more recently introduced platinum drugs. Although both inter- and intra-strand cross-links were effective, it was thought that in cells the inter-strand cross-link would, by preventing the separation of the strands necessary for cell division, and by being more difficult to repair, constitute the more effectively lethal lesion. With repair-deficient bacteria, it also emerged that a single cross-link in the genome was lethal, but proficient bacteria could remove about 20 cross-links through excision repair. Mono-7-alkylguanines were not removed and were evidently inert. Thus, only a few percent of the total alkylation products were the most effective lesions. Parallel studies with cultured mammalian cells gave a rather different picture, in that the mean lethal doses of even hypersensitive cell lines were around 20 or more cross-links per genome, about the same as for resistant strains of bacteria. Most cells could withstand several hundreds of cross-links per genome, and although adducts were removed, there was incomplete removal of cross-links. Some, but not all, sensitive cell lines were deficient in excision repair. Methods were devised for measuring the extents of alkylation of DNA in cells of patients treated with chemotherapeutic drugs; these are mainly immunoassays, and were applied generally to peripheral blood leukocytes, although some tumours were studied. Extents of alkylation of leukocyte DNA were generally of the same order as, or rather less than the mean lethal doses of cultured cells of the 'normal' type, but in some reports for cisplatin-treated patients, very wide variability between individuals was found. A positive correlation between adduct levels, and particularly a very minor adduct recognised specifically by one antibody, and favourable therapeutic outcome was discerned, and suggested to have a pharmacogenetic basis. In several instances, extents of alkylation of tumours were significantly higher than the average for leukocytes; for ovarian and a testicular tumour for cisplatin, and for a plasma cell tumour for melphalan. Nevertheless, these favourable examples would not constitute more than three or four mean lethal doses in the tumour cells, assuming that they had the same sensitivity as 'normal' cell lines: the therapeutic effect would of course be much more favourable if the tumour cells resembled 'sensitive' cell lines. This lack of a favourable difference between extents of alkylation in DNA of patients and the mean lethal dose for normal cells was particularly obvious with the methylating drugs dacarbazine and procarbazine. These considerations stress the need for higher extents of alkylation to be achieved in target tumour DNA for successful chemotherapy. One approach is to give a higher overall dose, and to 'rescue' the bone marrow (known from the earliest report on mustard gas to be the most susceptible tissue) by autologous transplantation. The second, which has yet to reach the clinic, is to convert unreactive prodrugs through enzymic activation into alkylating agents specifically in tumours (see Bagshawe, 1994).