Sources of discrepancy between the Hamilton Rating Scale for Depression (HRSD) and the Beck Depression Inventory (BDI) were investigated in 114 depressed inpatients treated with electroconvulsive therapy. Three previously reported observations were found to occur within the same sample: (a) There was only a moderate baseline correlation between the measures; (b) this correlation improved markedly at later assessment; (c) the HRSD had a greater effect size for change. The modest baseline correlation was largely due to patients who rated themselves as substantially less depressed than clinicians had rated them. Improvement in the correlation with repeated assessment was due to the representation of clinical responders. The larger effect size with the HRSD was mainly attributable to a small subgroup that showed marked deterioration on the BDI, a phenomenon not observed with clinician evaluation.