Forty to 50% of depressed patients who are initially prescribed antidepressant medications or administered electroconvulsive therapy do not experience a timely remission. This group typifies treatment‐refractory depression (TRD), defined as a failure to demonstrate an “adequate” response to an “adequate” treatment trial (i.e., sufficient intensity of treatment for sufficient duration). The approach to the patient with TRD must be methodical. The clinician should examine potential factors contributing to apparent non‐response: trial adequacy, compliance, differential diagnosis, and treatable comorbid conditions. After addressing these variables, a patient who does not demonstrate a remission may be considered treatment resistant (relative or absolute). Although many of these patients will respond to a subsequent treatment regimen, there are no (or only nominally useful) predictors for the initial selection of that “subsequent” antidepressant treatment. Hence, the initial treatment is typically chosen on the basis of safety and convenience, not differential efficacy. The search for the clinical and biological correlates of long‐term or acute outcome presents a major nosologic conundrum: Who will respond to treatment? Which treatment? In this manner, TRD challenges the prognostic utility of our current phenomenologic‐based diagnostic system. Depression and Anxiety 5:154–164, 1997. © 1997 Wiley‐Liss, Inc.