In clinical trials of antidepressant treatments, a depression rating score is usually measured at several points of time for each patient. We propose an approach to fit data from this type of clinical trial using an exponential mixed-effects model. Compared to previous proposals, this approach has the advantage that individual recovery curves are fitted rather than curves of means. Furthermore, no artificial fixing of model parameters is needed as in other approaches. The flexibility of the proposed model is shown for various situations. The approach is illustrated by an example from a placebo-controlled study for the treatment of depression with St. John's Wort (Hypericum perforatum).