We examine project drop-out in 39 sessions of psychodynamic group therapy. A total of 327 patients were included in the project, but 25.4% (83) did not evaluate outcome. According to therapists’ evaluation only 25.3% of the project drop-outs “improved” in “symptoms and problems” as compared with 79.1% of the project responders. According to Jacobson & Truax’s classification of Reliable Change, 52.9% of the project responders had a reliable improvement in the SCL-90 R Global Severity Index (GSI). Based on the relationship between therapist evaluations and GSI we estimated that only 43.7% of the whole samples might have had a reliable improvement in GSI. The SPSS standard statistical imputations procedure estimated that 48.6% of the patients reliably improved in GSI, and 50.2% when therapist evaluations were not included. It is concluded that therapist evaluations are essential in order to avoid bias in reported outcome solely based on project responders in the present naturalistic study, where outcome data probably are missing not at random.