This paper is based on data from the first Swiss Health Survey, but deals with a more general problem; the estimation of age of onset in illegal substance consumption. First, it is shown that taking the arithmetic mean for different age-groups as estimator leads to an underestimation of age of onset in the younger groups. Consequently, an alternative method is proposed based on survival models with right censoring. The prediction of age of onset is modeled with nonlinear functions. Results show for cannabis consumption in Switzerland that age of onset did not decrease notably during the past decade.