The literature on assessment of abuse liability of opioid drugs in humans was reviewed, and meta-analysis (a statistical method designed to combine results of different studies) was performed to study a dose-effect relationship for some variables commonly used for the evaluation of the subjective and physiological effects of morphine. Thirty-five studies, published between 1964 and 1991, including the i.m. or s.c. administration of morphine to non-dependent subjects with prior history of opioid abuse, were selected. Logistic regression models were applied for categorically dependent variables, and linear regression models were used for quantitative dependent variables. The log-transformation of the dose of morphine, stratifying by studies, acted as predictor variable. A dose-effect function was demonstrated. Doses producing an average effect across studies ranged from 7 mg (observers-reported 'liking') to 12 mg (subjects-reported 'liking'). Average effect on pupil diameter decreases was obtained for a dose of 8 mg. Doses producing a 50% of maximum positive responses were 10 mg for subjects' opioid classification, and 20 mg for the ARCI-MBG scale. The heterogeneity between studies obtained, along with the large variability present in most of the variables, supports the usefulness of a concurrent assessment of several indexes when assessing the effects of morphine.