OBJECTIVES:   Aim to determine the most cost-effective threshold of depression severity above which to prescribe anti-depressants to patients, in England and Wales, presenting with depression and under consideration for anti-depressants. Also aim to evaluate  the cost-effectiveness of a new trial of  anti-depressants in a population of wider range of depression severity than in previous trials. METHODS: Meta-regression of existing study results to estimate a proportional treatment effect on depression severity, which is then extrapolated to a wider range of severity than in included trials.  An economic model  which consists of a continuous outcome for the  initial 12 weeks of treatment, followed by a Markov model with states for depression category  and treatment. Treatment effects on Hamilton Depression Rating Scale (HAMD) were mapped to EQ5D. Expected value of partial perfect information (EVPPI) was used to determine an upper bound on the value of collecting ffurther evidence. RESULTS: Patients on anti-depressants had an additional 12% (CrI 3-21%) decrease in 6-week HAMD versus placebo. Treating patients with a severity  >2 on HAMD had the highest probability (>65%) of being cost-effective at £20,000 willingness-to-pay threshold. However this assumes that the relationship with severity can be extrapolated beyond the range of HAMD included in the systematic review. A short-term trial investigating the relation between treatment effect and severity and quality of life in depression patients had EVPPI=£67.7 million over a 10 year time-horizon.   CONCLUSIONS: On the basis of available evidence, our model suggests it may be cost-effective to treat patients with lower severity of symptoms than have been included in the majority of existing RCTs.  However, evidence of treatment efficacy in those with low HAMD is lacking, and there is likely to be value in conducting a trial on this lower severity population.