In this article we summarize the theoretical arguments which led us in the German Hodgkin's Disease Study Group to introduce the BEACOPP-regimen and to initiate a large randomised trial to investigate the role of moderate dose escalation in the treatment of advanced stage Hodgkin's disease. Although some indications for a role of dose were available in the early 1990s no prospective randomised trial had been undertaken. In order to obtain an impression of the shape of the essential dose response characteristic we developed a novel statistical model that could be used to analyse a set of data in which dose variations had occurred. The model took tumour growth and chemotherapy effects into account. The model could be applied to clinical data on tumour control and treatment given in a patient population. The model was fitted to the data of 706 patients which had received COPP/ABVD-like regimens. It revealed considerable heterogeneity in chemosensitivity and a positive slope of the doseresponse relationship. The model was used to simulate the effect of various treatment strategies with dose escalation and schedule changes. On the basis of such simulations we predicted that shortening cycle intervals from 4 to 3 weeks should lead to small benefits (about 3% in five-year tumour control rates). In contrast, we predicted that a moderate average dose escalation by 30% of a standard chemotherapy would lead to a potential benefit in the order of 10%-15% in tumour control at five years. Subsequently we searched for a treatment scheme that would permit such a dose escalation. The BEACOPP-scheme was invented to allow the three major myelotoxic substances (cyclophsphamide, adriamycin and etoposide) to be given in the beginning of a cycle. These three substances were then subject to dose escalation in a dose finding trial. G-CSF was introduced to compensate for the myelotoxic effects. The dose level found feasible for a large multicentre setting turned out to be in the required magnitude. The HD9 trial of the GHSG was then initiated to examine whether the predicted dose response curve really exists.