Our ability to predict the outcome of psychological treatments, particularly for drug dependence, is examined by (1) new data on a VA sample, (2) a review of studies predicting the outcome of drug abuse treatment, and (3) a review of predicting the outcome of psychotherapy for other types of patients. For (1), the direct predictions of 13 staff members' ratings of the outcome of treatment correlated. 27 with the outcome. Although staff predictions improved when grosser predictions were examined, the results are disappointing given the level of discrimination required and the modest levels of prediction attained. For (2), prediction success with drug abuse patients in other studies were also disappointing and the bases for predictions often did not hold up on crossvalidation. For (3), direct predictions in the Penn Psychotherapy Study were similar to the VA sample in level of success. Other studies produced insignificant or similarly low levels of prediction success. In our discussion we suggest that rather than relying mainly upon pretreatment status for prediction, two promising areas should be examined in future studies: (1) the patient's early response to the treatment and to the therapist, (2) the patient's environment posttreatment that should be altered to consolidate and perpetuate the gains made during treatment. In conclusion, it is true that we are surprisingly unable to predict the outcomes of psychological treatments beyond a very modest level.