Summary In order to assess the time at which the distant metastases were initiated, a model has been developed to simulate the natural history of human breast cancer. The metastasis appearance curves were fitted to those observed for tumours of various sizes among the 2648 patients treated at the Institut Gustave Roussy from 1954 to 1972. The model assumes that metastases are initiated when the tumour reaches a threshold volume (distribution of this volume was estimated in a previous article). Two With exponential growth, it was necessary to introduce correlations to obtain a satisfactory fit of the metastases appearance curves: between the tumour volume at diagnosis and the doubling time (RI = -0.3), and between the tumour volume at metastasis initiation and the doubling time (R2=0.3). The growth duration of the metastases before their detection was found to equal about 18 metastases doubling times at detection and the mean ratio between the doubling time of a tumour and its metastases equal to 2.2.With Gompertzian growth, it was impossible to adjust satisfactorily the proportions of metastases at diagnosis as a function of the primary tumour volume. However, when we ignore this, the best fit was obtained when the duration of metastases growth before detection was about the same as for exponential growth.With either growth pattern, the model predicts that the proportion of patients with metastases would be reduced by -30% if the primary tumours were treated 12 months earlier. This prediction is consistent with the results of the screening programs for breast cancer.In a previous article we studied the relationship between the size of the primary tumour at initial treatment and the incidence of distant metastases during the course of the disease. The data showed that the primary tumour at the time of metastasis initiation is, on the average, only slightly smaller than at the time of tumour diagnosis (Koscielny et al., 1984).This conclusion conflicts with the predictions of previous models of the natural history of breast cancer. Igot & Legal (1968), and later Breuer (1976) assumed that metastases grow exponentially from the time of origin and they extrapolated metastases growth curves backwards in order to determine when the metastases contained one single cell. Assuming identical growth rates for both primary tumours and metastases, they concluded that metastasis commences very early in the development of the primary.This divergence can be explained by the following hypotheses: (a) the number of cells necessary for Correspondence: S. Koscielny. Received 1 January 1985; and in revised form, 11 June 1985. metastasis initiation is much greater than one. (b) the metastasis growth rate is more rapid than the primary tumour one. (c) tumour growth is not exponential.The model proposed in this paper was developed in order to assess the time at which the distant metastases were initiated, taking into account the clinical data, particularly the relationship between primary tumour size and the metastases appearanc...